New approaches in intelligent image analysis techniques, methodologies and applications

Volume 108 Intelligent Systems Reference Library Series Editors Janusz Kacprzyk Polish Academy of Sciences, Systems Research Institute, Warsaw, Poland Lakhmi C Jain Bournemouth University, University of Canberra and, ACT, Aust Capital Terr, Australia About this Series The aim of this series is to publish a Reference Library, including novel advances and developments in all aspects of Intelligent Systems in an easily accessible and well structured form The series includes reference works, handbooks, compendia, textbooks, well-structured monographs, dictionaries, and encyclopedias It contains well integrated knowledge and current information in the field of Intelligent Systems The series covers the theory, applications, and design methods of Intelligent Systems Virtually all disciplines such as engineering, computer science, avionics, business, e-commerce, environment, healthcare, physics and life science are included More information about this series at http://​www.​springer.​com/​series/​8578 Editors Roumen Kountchev and Kazumi Nakamatsu New Approaches in Intelligent Image Analysis Techniques, Methodologies and Applications Editors Roumen Kountchev Department of Radio Communications and Video Technologies, Technical University of Sofia, Sofia, Bulgaria Kazumi Nakamatsu School of Human Science and Environment, University of Hyogo, Himeji, Japan ISSN 1868-4394 e-ISSN 1868-4408 ISBN 978-3-319-32190-5 e-ISBN 978-3-319-32192-9 DOI 10.1007/978-3-319-32192-9 Library of Congress Control Number: 2016936421 © Springer International Publishing Switzerland 2016 This work is subject to copyright All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed The use of general descriptive names, registered names, trademarks, service marks, etc in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made Printed on acid-free paper This Springer imprint is published by Springer Nature The registered company is Springer International Publishing AG Switzerland Preface This book represents the advances in the development of new approaches, used for the intelligent image analysis It introduces various aspects of the image analysis, related to the theory for their processing, and to some practical applications The book comprises 11 chapters, whose authors are researchers from different countries: USA, Russia, Bulgaria, Japan, Brazil, Romania, Ukraine, and Egypt Each chapter is a small monograph, which represents the recent research work of the authors in the corresponding scientific area The object of the investigation is new methods, algorithms, and models, aimed at the intelligent analysis of signals and images—single and sequences of various kinds: natural, medical, multispectral, multiview, sound pictures, acoustic maps of sources, etc New Approaches for Hierarchical Image Decomposition, Based on IDP, SVD, PCA, and KPCA In Chap the basic methods for hierarchical decomposition of grayscale and color images, and of sequences of correlated images are analyzed New approaches are introduced for hierarchical image decomposition: the Branched Inverse Difference Pyramid (BIDP) and the Hierarchical Singular Value Decomposition (HSVD) with tree-like computational structure for single images; the Hierarchical Adaptive Principle Component Analysis (HAPCA) for groups of correlated images and the Hierarchical Adaptive Kernel Principal Component Analysis (HAKPCA) for color images In the chapter the evaluation of the computational complexity of the algorithms used for the implementation of these decompositions is also given The basic application areas are defined for efficient image hierarchical decomposition, such as visual information redundancy reduction; noise filtration; color segmentation; image retrieval; image fusion; dimensionality reduction, where the following is executed: the objects classification; search enhancement in large-scale image databases, etc Intelligent Digital Signal Processing and Feature Extraction Methods The goal of Chap is to present well-known signal processing methods and the way they can be combined with intelligent systems in order to create powerful feature extraction techniques In order to achieve this, several case studies are presented to illustrate the power of hybrid systems The main emphasis is on the instantaneous time–frequency analysis, since it is proven to be a powerful method in several technical and scientific areas The oldest and most utilized method is the Fourier transform, which has been applied in several domains of data processing, but it has very strong limitations due to the constraints it imposes on the analyzed data Then the short-time Fourier transform and the wavelet transform are presented as they provide both temporal and frequency information as opposed to the Fourier transform These methods form the basis of most applications, as they offer the possibility of time–frequency analysis of signals The Hilbert–Huang transform is presented as a novel signal processing method, which introduces the concept of the instantaneous frequency that can be determined for every time point, making it possible to have a deeper look into different phenomena Several applications are presented where fuzzy classifiers, support vector machines, and artificial neural networks are used for decision-making Interconnecting these intelligent methods with signal processing will result in hybrid intelligent systems capable of solving computationally difficult problems Multi-dimensional Data Clustering and Visualization via Echo State Networks Chapter summarizes the proposed recently approach for multidimensional data clustering and visualization It uses a special kind of recurrent networks called Echo State Networks (ESN) to generate multiple 2D projections of the multidimensional original data The 2D projections are subjected to selection based on different criteria depending on the aim of particular clustering task to be solved The selected projections are used to cluster and/or to visualize the original data set Several examples demonstrate the possible ways to apply the proposed approach to variety of multidimensional data sets: steel alloys discrimination by their composition; Earth cover classification from hyperspectral satellite images; working regimes classification of an industrial plant using data from multiple measurements; discrimination of patterns of random dot motion on the screen; and clustering and visualization of static and dynamic “sound pictures” by multiple randomly placed microphones Unsupervised Clustering of Natural Images in Automatic Image Annotation Systems Chapter is devoted to automatic annotation of natural images joining the strengths of the text-based and the content-based image retrieval The automatic image annotation is based on the semantic concept models, which are built from large number of patches received from a set of images In this case, image retrieval is implemented by keywords called Visual Words (VWs) that is similar to text document retrieval The task involves two main stages: a low-level segmentation based on color, texture, and fractal descriptors and a high-level clustering of received descriptors into the separated clusters corresponding to the VWs set The enhanced region descriptor including color, texture, and fractal features has been proposed For the VWs generation, the unsupervised clustering is a suitable approach The Enhanced Self-Organizing Incremental Neural Network (ESOINN) was chosen due to its main benefits as a self-organizing structure and online implementation The preliminary image segmentation permitted to change a sequential order of descriptors entering the ESOINN as associated sets Such approach simplified, accelerated, and decreased the stochastic variations of the ESOINN The experiments demonstrate acceptable results of the VWs clustering for a non-large natural image sets This approach shows better precision values and execution time as compared to the fuzzy c-means algorithm and the classic ESOINN Also issues of parallel implementation of unsupervised segmentation in OpenMP and Intel Cilk Plus environments were considered for processing of HD-quality images An Evolutionary Optimization Control System for Remote Sensing Image Processing Chapter provides an evolutionary control system via two Darwinian Particle Swarm Optimizations (DPSO)—one novel application of DPSO—coupled with remote sensing image processing to help in the image data analysis The remote sensing image analysis has been a topic of ongoing research for many years and has led to paradigm shifts in the areas of resource management and global biophysical monitoring Due to distortions caused by variations in signal/image capture and environmental changes, there is not a definite model for image processing tasks in remote sensing and such tasks are traditionally approached on a case-by-case basis Intelligent control, however, can streamline some of the case-by-case scenarios and allows faster, more accurate image processing to support the more accurate remote sensing image analysis Tissue Segmentation Methods Using 2D Histogram Matching in a Sequence of MR Brain Images In Chap a new transductive learning method for tissue segmentation using a 2D histogram modification, applied to Magnetic Resonance (MR) image sequence, is introduced The 2D histogram is produced from a normalized sum of co-occurrence matrices of each MR image Two types of model 2D histograms are constructed for each subsequence: intra-tissue 2D histogram to separate tissue regions and an inter-tissue edge 2D histogram First, the MR image sequence is divided into few subsequences, using wave hedges distance between the 2D histograms of the consecutive MR images The test 2D histogram segments are modified in the confidence interval and the most representative entries for each tissue are extracted, which are used for the kNN classification after distance learning The modification is applied by using LUT and two ways of distance metric learning: large margin nearest neighbor and neighborhood component analysis Finally, segmentation of the test MR image is performed using back projection with majority vote between the probability maps of each tissue region, where the inter-tissue edge entries are added with equal weights to corresponding tissues The proposed algorithm has been evaluated with free access data sets and has showed results that are comparable to the state-of-the-art segmentation algorithms, although it does not consider specific shape and ridges of brain tissues Multistage Approach for Simple Kidney Cysts Segmentation in CT Images In Chap a multistage approach for segmentation of medical objects in Computed Tomography (CT) images is presented Noise reduction with consecutive applied median filter and wavelet shrinkage packet decomposition, and contrast enhancement based on Contrast limited Adaptive Histogram Equalization (CLAHE) are applied in the preprocessing stage As a next step a combination of two basic methods is used for image segmentation such as the split and merge algorithm, followed by the color-based K-mean clustering For refining the boundaries of the detected objects, additional texture analysis is introduced based on the limited Haralick’s feature set and morphological filters Due to the diminished number of components for the feature vectors, the speed of the segmentation stage is higher than that for the full feature set Some experimental results are presented, obtained by computer simulation The experimental results give detailed information about the detected simple renal cysts and their boundaries in the axial plane of the CT images The proposed approach can be used in real time for precise diagnosis or in disease progression monitoring Audio Visual Attention Models in Mobile Robots Navigation In Chap , it is proposed to use the exiting definitions and models for human audio and visual attention, adapting them to the models of mobile robots audio and visual attention, and combining with the results from mobile robots audio and visual perception in the mobile robots navigation tasks The mobile robots are equipped with sensitive audio visual sensors (usually microphone arrays and video cameras) They are the main sources of audio and visual information to perform suitable mobile robots navigation tasks modeling human audio and visual perception The audio and visual perception algorithms are widely used, separately or in audio visual perception, in mobile robot navigation, for example to control mobile robots motion in applications like people and objects tracking, surveillance systems, etc The effectiveness and precision of the audio and visual perception methods in mobile robots navigation can be enhanced combining audio and visual perception with audio and visual attention There exists relative sufficient knowledge describing the phenomena of human audio and visual attention Local Adaptive Image Processing Three methods for 2D local adaptive image processing are presented in Chap In the first one, the adaptation is based on the local information from the four neighborhood pixels of the processed image and the interpolation type is changed to zero or bilinear The analysis of the local characteristics of images in small areas is presented, from which the optimal selection of thresholds for dividing into homogeneous and contour blocks is made and the interpolation type is changed adaptively In the second one, the adaptive image halftoning is based on the generalized 2D Last Mean Square (LMS) error-diffusion filter for image quantization The thresholds for comparing the input image levels are calculated from the gray values dividing the normalized histogram of the input halftone image into equal parts In the third one, the adaptive line prediction is based on the 2D LMS adaptation of coefficients of the linear prediction filter for image coding An analysis of properties of 2D LMS filters in different directions was made The principal block schemes of the developed algorithms are presented An evaluation of the quality of the processed images was made on the base of the calculated objective criteria and the subjective observation The given experimental results, from the simulation for each of the developed algorithms, suggest that the effective use of local information contributes to minimize the processing error The methods are suitable for different types of images (fingerprints, contour images, cartoons, medical signals, etc.) The developed algorithms have low computational complexity and are suitable for real-time applications Machine Learning Techniques for Intelligent Access Control In Chap 10 several biometric techniques, their usage, advantages and disadvantages are introduced The access control is the set of regulations used to access certain areas or information By access we mean entering a specific area, or logging on a machine (PC, or another device) The access regulated by a set of rules that specifies who is allowed to get access, and what are the restrictions on such access Over the years several basic kinds of access control systems have been developed With advancement of technology, older systems are now easily bypassed with several methods, thus the need to have new methods of access control Biometrics is referred to as an authentication technique that relies on a computer system to electronically validate a measurable biological characteristic that is physically unique and cannot be duplicated Biometrics has been used for ages as an access control security system Experimental Evaluation of Opportunity to Improve the Resolution of the Acoustic Maps Chapter 11 is devoted to generation of acoustic maps The experimental work considers the possibility to increase the maps resolution The work uses 2D microphone array with randomly spaced elements to generate acoustic maps of sources located in its near-field region In this region the wave front is not flat and the phase of the input signals depends on the arrival direction, and on the range as well The input signals are partially distorted by the indoor multipath propagation and the related interference of sources emissions For acoustic mapping with improved resolution an algorithm in the frequency domain is proposed The algorithm is based on the modified method of Capon Acoustic maps of point-like noise sources are generated The maps are compared with the maps generated using other famous methods including built-in equipment software The obtained results are valuable in the estimation of direction of arrival for Noise Exposure Monitoring This book will be very useful for students and Ph.D students, researchers, and software developers, working in the area of digital analysis and recognition of multidimensional signals and images Roumen Kountchev Kazumi Nakamatsu Sofia, Bulgaria, Himeji, Japan 2015 Contents New Approaches for Hierarchical Image Decomposition, Based on IDP, SVD, PCA and KPCA Roumen Kountchev and Roumiana Kountcheva 1.​1 Introduction 1.​2 Related Work 1.​3 Image Representation Based on Branched Inverse Difference Pyramid 1.​3.​1 Principles for Building the Inverse Difference Pyramid 1.​3.​2 Mathematical Representation of n-Level IDP 1.​3.​3 Reduced Inverse Difference Pyramid 1.​3.​4 Main Principle for Branched IDP Building 1.​3.​5 Mathematical Representation for One BIDP Branch 1.​3.​6 Transformation of the Retained Coefficients into Sub-blocks of Size × 1.​3.​7 Experimental Results 1.​4 Hierarchical Singular Value Image Decomposition 1.​4.​1 SVD Algorithm for Matrix Decomposition 1.​4.​2 Particular Case of the SVD for Image Block of Size × 1.4.3 Hierarchical SVD for a Matrix of Size n × n 1.4.4 Computational Complexity of the Hierarchical SVD of Size n × n 1.​4.​5 Representation of the HSVD Algorithm Through Tree-like Structure 1.​5 Hierarchical Adaptive Principal Component Analysis for Image Sequences 1.​5.​1 Principle for Decorrelation of Image Sequences by Hierarchical Adaptive PCA 1.​5.​2 Description of the Hierarchical Adaptive PCA Algorithm 1.​5.​3 Setting the Number of the Levels and the Structure of the HAPCA Algorithm signals of channels ADC digitizes those signals with sampling interval , which acts as the time delay increment For instance, if the sound source emits monochromatic input signal with frequency 10 kHz digitized by ADC with sampling frequency 65.536 kHz One period of the signal will have ADC counts The latest corresponds to the time delay increment of The big increment value limits available angular positions of the microphone array beam, affects its sidelobe level and its pattern performance The focalization in frequency domain enables to alleviate such limitation A channel output signal obtained during the observation time is denoted as a column with samples (11.4) We obtain spectrum realization for this column by its Fourier transform We denote by matrix all the spectrums realizations obtained; the matrix consists of rows and columns ( ) For each of rows, we calculate the matrix, using the product of a row conjugate transposed and the row We obtain the realization of the cross-spectrum matrix as a result of the sum of all matrices We denote by matrix the realization of cross-spectrum matrix obtained for defined bandwidth The acoustic camera enables operation in wide range of frequencies from 10 to 25.6k Hz The acoustic imaging approach requires defining center frequency and bandwidth around it The work considers narrowband signals those bandwidth is not more than 10 % of the center frequency The time delays are calculated for the defined range of the acoustic map grid The phase delays are calculated for the defined: (a) acoustic signal center frequency ; (b) the grid nodes coordinates The microphone array spatial (angular) scan delays are regarded as rows with elements equal to: The beam scanning is realized via adjustment of delays of the row with respect to the defined range of the grid and beam angular position The delay-and-sum (DAS) is the mostly used beamforming method based on the sum of signals with respect to considered delays [15] The matrix form representation of the method can be figured out in the frequency domain Taking into account all considerations mention above, the output signal power , in the particular node of acoustic map, is equal to: (11.5) where superscript letter denotes conjugate transpose The internal noises of microphones, preamplifiers, and the ADC dither are considered as the noncorrelated additive noise of channels of acoustic camera equipment The noise main contribution is concentrated in the main diagonal elements of the cross-spectrum matrix [15] The matrix diagonal elements removal increases the signal-to-noise ratio of generated acoustic maps, suppresses sidelobe level up to 0.6 dB [15] We denote cross-spectrum matrix with nulled elements of main diagonal by Thus, the power of the output signal , in the particular node of acoustic map, becomes: (11.6) A method for high-resolution spectrum analysis was proposed in 1969 by Capon [8] Later the Capon method was described for estimation of the direction of arrival of a signal [11] The method does not need any prior information about the number of sources in the field of view [11] The method does not require prior knowledge about signal amplitude distribution The method is limited by requirements to inversion of matrices and to signal-to-noise ratio more than 10 dB The method applicability is extended by: (a) addition unit matrix of size to the matrix under consideration; (b) modern methods of matrix pseudo-inverse Signal processing can be done in the frequency domain We use the modified pseudo inverse cross-spectrum matrix , which calculation includes the unit matrix addition The output signal power map is equal to: in a particular node of acoustic (11.7) 11.3 The Experimental Acoustic Camera Equipment The acoustic camera is suitable for frequency, time and spatial analysis The acoustic camera utilizes a fusion of optical image and acoustic map Brüel&Kjaer (B&K) Sound and Vibration Measurement A/S manufactured both camera software and hardware The camera equipment includes the microphone array with the optical camera; 6- and 12-channel input modules and the laptop with software for acoustic signals analysis (Fig 11.3) The camera microphone array is B&K type WA1558-W-021 [16] It is two-dimensional randomly distributed microphone wheel array with diameter ≈0.33 m Fig 11.3 Block diagram of the acoustic camera The quantity of the microphones is (Fig 11.4) The input modules are B&K type 3053-B120 and type 3050-B-060 The signals used for acoustic map generation (11.5–11.7) are recorded using Time Data Recorder option of B&K Pulse LabShop software The input modules dynamic range is up to 160 dB; the modules provide high phase stability and interchannel isolation The above factors allow analysis of weak acoustic emissions Fig 11.4 The layout of microphones in the array The acoustic camera microphones B&K type 4958 have built-in preamplifiers and contain transducer electronic data sheets These datasheets are aimed to transfer each microphone features to input modules The microphones dimensions are: 34 mm × mm Their sensitivity is 11.2 mV/Pa Their operation temperatures range is from −10 to +55 °C Their dynamic range is from 28 to 140 dB The quality of acoustic signal amplitude estimation is provided by the acoustic camera amplitude calibration The pistonphone calibrator B&K type 4228 corresponds to requirements of IEC 942 (1988) Class 1L or Class 0L (with external barometer) and ANSI S1.40-1984 The calibrator has the high stability of calibration signal level and frequency The calibration signal accuracy is provided in a wide range of operation temperature, humidity, and pressure The calibrator is battery-operated The frequency of the signal of the pistonphone calibrator is 251.2 Hz The calibrator adaptor DP0775 is suitable for sequential calibration of the acoustic camera channels The input module with extended dynamic range is 6-channel The second input module is 12channel Its resolution is 24 bit Its interchannel leakage is not worse than −80 dB according to its datasheet The input modules dynamic range depends on sampling frequency and a bandwidth These modules support transducer electronic data sheet in order to provide the B&K Response Equalization technique The modules are mounted in module frame B&K type 3660-C-000 for modules The battery module B&K type 2831 is mounted in the frame as well The microphones and input modules features are given in Table 11.1 Table 11.1 Frequency characteristics for modules of the Acoustic Camera Module Name Type Microphones 4958 Frequency range, kHz Features 0.01–20 – Input modules 3053-B-120 0–25.6 Sampling rate: 65.5 k samples/s Number of input channels: 12 3050-B-060 0–51.2 Sampling rate: 131 k samples/s Number of input channels: The acoustic camera input modules ADCs sampling frequency is 65.536 kHz The highest frequency of the camera channels is 25.6 kHz The sampling ratio equals to 2.56 The input modules output signals are synchronized according to IEEE 1588 Precision Time Protocol [17] The acoustic camera software includes: Acoustic Test Consultant type 7761; Beamforming type 8608; FFT Analysis type 7770; Time Data Recorder Type 7708 The software provides generation of acoustic maps in defined frequency range, time and frequency representation of signals, their recording, etc The signals can be stored for further post-processing with third-party software using multi-buffer from Time Data Recorder option of B&K Pulse LabShop software We use observation time equal to 0.25 s for each multi-buffer The signals are stored to the hard disk drive with time stamps and the acoustic camera settings The post-processing is done with developed scripts using MATLAB computing environment 11.4 Experimental Results The experiments on the estimation of angular coordinates of several point-like sources of acoustic signals were held in an office room The room is not optimized in terms of multipath propagation of acoustic signals inside it The above approach enables to compensate the narrowband signals delays including such from near-field region of the considered two-dimensional microphone array Two sources of acoustic signals are placed on the range of 0.78 m which is comparable to the microphone array dimensions The picture of the experimental setup is given on the Fig 11.5 Fig 11.5 Picture of imaging scenario: two point-like sources of the acoustic signal in front of the acoustic camera The amplitude calibration is performed after the Acoustic Camera switching-on The amplitude calibration uses the calibration option of B&K Pulse LabShop software and the pistonphone calibrator described above The source of the acoustic signal is placed on array boresight direction (PC speaker marked as “Source 1” in Fig 11.5) The source emits acoustic noise signal The sources of acoustic signals are × W stereo speakers; their switching-on increased the Acoustic Camera output signal (in the frequency range from 10 to 25.6 kHz) from 17 to 20 dB The experimental estimation of the two-dimensional microphone array pattern is done for several center frequencies The signal bandwidth is 10 % of its center frequency The chosen values of center frequencies belong to the frequency range from 0.1 to 18 kHz The observation time is not less than 0.25 s The acoustic signal source (“Source 1” in Fig 11.5) position is constant during the observation time The experiments are held for the microphone array field of view ±90° in azimuth and in elevation The microphone array pattern (AP) parameters under investigation are: (a) beamwidth of the AP; (b) the AP sidelobes levels (SLL); (c) the AP highest null level; (d) angular position of highest sidelobe of the AP The highest null and the highest SLL are considered as such with the highest values All further estimations are held using normalized AP 11.4.1 Microphone Array Patterns Generated with the Delay-and-Sum Beamforming Method The experimental estimation of the two-dimensional microphone array pattern is done using DAS beamforming (11.5) The AP beamwidth is measured as the normalized AP half-power level (−3 dB level) regarding its peak We denote the beamwidth as and in azimuth and in elevation, correspondingly, where superscript symbol denotes that the value is given in degrees For center frequencies lower than 500 Hz the estimated AP lowest level in the field of view not cross the −3 dB level thus, the AP beamwidth can not be estimated For center frequency 500 Hz the 141°; the elevation slice lowest level is −2.7 dB Estimated microphone array patterns parameters are given in Table 11.2 for center frequencies from to 18 kHz The angular position of the highest sidelobe of the AP varies from 85° to 13° The highest null level of the AP varies from −2.8 to −18 dB The highest null level estimated on center frequency kHz equals to −11 dB in azimuth and −12 dB in elevation Table 11.2 The parameters of the microphone array pattern estimated indoor using delay-and-sum beamforming method f C , kHz β° ε° Sidelobes levels, dB No No Az El Az El 61.5 59.3 −7.8 −8.2 – – 17.8 20.5 −8.4 −11 −11.5 −9.4 – 10 −8.5 −16 18 3.4 3.4 −12.2 −15.8 −11.6 −12.1 −6.6 −9.2 −6.6 −8 −11.5 −16 No No No Az El Az El Az El – – – – – – – – – – – −5.3 −7.8 −11.4 −8.7 −8.2 −9.8 −9.2 −10 The variation of the listed values along repeated measurements is inconsiderable The repeatability depends on the experiment conditions and on parameter stability of B&K equipment 11.4.2 Microphone Array Patterns Generated with the Christensen Beamforming Method The sidelobes of the array pattern determine the level of penetration of unwanted acoustic signals from corresponding angular directions It is known that the removal of main diagonal of the crossspectrum matrix can diminish the sidelobes level [3, 15] The considered above signals recordings were processed with Christensen beamforming method (11.6) The estimated AP parameters are given in Table 11.3 Table 11.3 The parameters of the microphone array pattern estimated indoor using Christensen beamforming method f C , kHz β° ε° Sidelobes levels, dB No No Az El Az El 56.6 57.8 −10.9 −11.9 – – Highest null level, dB 16.5 18.8 −10.4 −10.5 −10.7 −10.4 – 10 5.5 18 3.2 3.2 −12 No Az El – – – No Az El – – No Az El – – Az – El – – – −11 −10.5 – – −12.1 −11.3 −11.4 −11.2 −6.3 −10.4 −11.2 −12.5 −12.2 −11.4 −12.5 −12.1 −8 −11.8 −12.4 −12.3 −12.3 −12.4 −7.4 −12.4 −13 −12.2 −12.4 For center frequencies lower than 500 Hz the AP beamwidth is about 10° narrower than estimated one by DAS beamforming method For center frequency 500 Hz the 128° and the 152° Thus, Christensen beamforming method enabled to estimate for center frequency 500 Hz unlike the DAS beamforming method For center frequency kHz, the Christensen beamforming enabled to narrow the beamwidth in 4.9° and in 1.5° comparable to DAS beamforming (Table 11.2) Christensen paper is focused on sidelobe level suppression, not on the main lobe narrowing [3, 15] Let us compare the level of unwanted penetration via sidelobes in AP, estimated with DAS and Christensen beamforming methods (Tables 11.2, 11.3) The first sidelobe is considered For center frequency kHz, the sidelobe is suppressed on 3.1 and 3.7 dB compared to such levels estimated with DAS beamforming Such comparison for center frequency kHz shows suppression from to dB for the second sidelobe in elevation and the first sidelobe in azimuth, correspondingly The second sidelobe in azimuth is 0.8 dB higher and the first sidelobe in elevation is 0.5 dB higher For center frequency 10 kHz, the Christensen beamforming enables to suppress first and second sidelobes in azimuth from 0.6 to 2.9 dB but makes such sidelobes in elevation higher from 4.7 to 4.8 dB For center frequency 18 kHz, the first and the second sidelobes are higher from 0.2 to 3.7 dB The latest may be explained with indoor experimental conditions The difference of the AP beamwidth estimated with DAS and Christensen beamforming methods is given in Table 11.4 The beamwidth values depend on the wavelength (corresponding to center frequency) and the microphone array lengths in corresponding angular directions as: and as Let us calculate the values and for above-shown parameters: speed of sound m/s; center frequency kHz; the microphone array lengths m and m The calculated values are equal approximately to 6.1° and 5.9° in azimuth and elevation The values comparison to those from Tables 11.2, 11.3 and 11.4 shows that in the indoor experiment, the DAS beamforming delivers beamwidth 1° wider than calculated one and Christensen beamforming narrows it less than 1° Table 11.4 Beamwidth difference of the microphone array patterns estimated with delay-and-sum beamforming method and Christensen beamforming method for center frequency 10 kHz Parameter DAS Christensen Difference β° 5.5 2.5 ε° 11.4.3 Microphone Array Patterns Generated with the Modified CaponBased Beamforming Method The considered above recordings of signals were processed with modified Capon-based beamforming method (11.7) For center frequencies from 1, 3, 10 and 18 kHz, the estimated microphone array patterns parameters are given in Table 11.5 Table 11.5 The parameters of the microphone array pattern estimated indoor using modified Capon-based beamforming method f C , kHz β° ε° Sidelobes levels, dB From To 48.3 49.3 −5.3 −3.8 16.2 18.8 −5.1 −4.1 10 2.8 2.8 −7.4 −6.8 18 1.25 1.3 −11.5 −10 For the lowest center frequency 59° and 69° The time-bandwidth product of the narrowband noise signal with this center frequency is about dB [18] The equipment instability in the experiment rejects to obtain such result in long-time observations Other methods did not enable to estimate beamwidth on the center frequency 100 Hz in the defined above field of view For the center frequency 500 Hz, the estimated beamwidth equals to 50° and 59° The latest values are 2.56 times narrower comparable to such estimated using Christensen method; the AP generated with DAS beamforming method does not cross the −3 dB level in the ±90° field of view, as it was noted above The obtained sidelobes level and highest null level are equal to approximately −7.4 dB For center frequency 10 kHz, the peak sidelobe level is approximately −6.8 dB Its angular position corresponds to the estimated with DAS and Christensen beamforming methods The position in azimuth is approximately 30° Let us compare the unwanted penetration level estimated with the modified Capon-based and DAS beamforming methods The first and the second sidelobes are under consideration For center frequencies 1, 3, and 10 kHz, the Capon sidelobe level is higher For center frequency kHz it is higher from 2.5 to 4.4 dB; for center frequency kHz it is higher from 3.3 to 7.4 dB; for center frequency 10 kHz it is higher from 1.1 to 9.2 dB, the latest can be affected by the existing multipath propagation of the acoustic signal inside the office room The difference in sidelobe levels those obtained using Capon-based and DAS beamforming methods varies from 5.8 to −0.1 dB, for center frequency 18 kHz These level lower values were obtained in the third sidelobe angular position for center frequencies 10 and 18 kHz The improvement is from 1.5 to 3.4 dB Let us compare the unwanted penetration level estimated with the modified Capon-based and Christensen beamforming methods The first and the second sidelobes are again under consideration For center frequencies from to 10 kHz, the Capon level is higher For center frequency kHz it is higher from 5.6 to 8.1 dB; for center frequency kHz it is higher from 5.3 to 6.6 dB; for center frequency 10 kHz it is higher from 3.8 to 5.3 dB For center frequency 18 kHz, the level can be 3.5 dB lower as well as 2.1 dB higher The 3.5 dB improvement is obtained in the second sidelobe in azimuth The difference of the AP beamwidth estimated with the modified Capon-based beamforming method to such estimated with DAS and Christensen beamforming methods is given in Table 11.6 In the table, the ratio between beamwidth values those estimated with DAS and modified Capon-based beamforming methods is given in column The column shows the ratio between beamwidth values estimated with Christensen and modified Capon-based beamforming methods The insignificant mainlobe width narrowing for center frequency kHz ( in Table 11.6) is connected to indoor conditions of the experiment Table 11.6 Beamwidth ratio of such estimated with delay-and-sum and Christensen beamforming methods to values estimated using modified Capon-based beamforming method f C , kHz k k2 β° ε° β° ε° 1.27 1.2 1.17 1.17 1.09 1.09 1.02 10 2.5 2.5 1.96 1.79 18 2.72 2.62 2.56 2.46 In the experimental study usage of the modified Capon-based beamforming method (11.7) instead the DAS beamforming method (11.5) enabled to narrow the mainlobe width from 1.17 to 1.27 times for center frequency kHz; from 1.79 to 2.5 times for center frequency 10 kHz; from 2.46 to 2.72 times for center frequency 18 kHz (Table 11.6) The mainlobe width narrowing obtained by using modified Capon-based beamforming method instead of the Christensen beamforming method is from 1.17 to 2.56 times (Table 11.6) The results obtained in the center frequency range from 100 Hz up to 500 Hz show that acoustic maps generation in the field of view ±90° is suitable using modified Capon-based beamforming method only The obtained AP beamwidth narrowing increases the microphone array angular resolution that improves the generated acoustic map quality 11.4.4 Microphone Array Responses for Two Point-like Emitters The acoustic maps of two point-like sources (Fig 11.5) of acoustic noise signals were generated The signal bandwidth, the signal source range, and other experiment parameters are similar to previous ones The signals center frequency is 10 kHz The sources are separated is azimuth plane approximately on 0.085 m In the defined range, their azimuth angular separation is The latest equals approximately to the calculated AP beamwidth and is less than the beamwidth estimated using DAS beamforming method (Tables 11.2 and 11.3) For further experiments the acoustic maps field of view is equal to ±35° in both azimuth and elevation directions; the acoustic map slices field of view is equal to ±90° in both angular directions The power of the acoustic signal sources is approximately equal Acoustic maps normalized are under consideration These maps threshold is −3 dB That means that the map nodes with power less than dB comparable to the map peak are not shown The normalized map slices have a dashed line for −3 dB level The acoustic maps are generated with DAS (11.5), Christensen (11.6) and modified Capon-based beamforming methods (11.7) The acoustic map generated by modified Capon-based beamforming method (11.7) is shown in Fig 11.6 The level of the hollow between the two peaks approximately equals to −3 dB, so they will be resolved as two sources The peak sidelobe level is approximately equal to −4 dB For the center frequency, the grating lobes position in azimuth is approximately equal to ±30° as was noted above Fig 11.6 Acoustic map generated using the modified Capon beamforming method: a acoustic map; b slice of the map for elevation 1, deg The heuristic, modified methods, like squaring of the method (11.7) enable to improve the obtained result The noted peak sidelobe level can be suppressed to about −8 dB (Fig 11.7) Fig 11.7 Acoustic map generated using the heuristic method—squared modified Capon-based beamforming method: a acoustic map; b slice of the map for elevation 1, deg The acoustic map generated by Christensen beamforming method (11.6) is shown in Fig 11.8 The same signal recordings were used to generate this map The level of the hollow between peaks in Fig 11.8 is lower than −3 dB These peaks cannot be resolved The unwanted penetration level is lower than such for modified Capon-based beamforming method Fig 11.8 Acoustic map generated using the Christensen beamforming method: a acoustic map; b slice of the map for elevation 1, deg The acoustic map generated by DAS beamforming method (11.5) is shown in Fig 11.9 The sidelobes level rise to −3 dB level (dashed line) The map shows one peak only The map responses at some angular coordinates from 20° to 40° correspond to acoustic signal multipath propagation inside the office room and to sidelobes level of the microphone array as well Fig 11.9 Acoustic map generated using the delay-and-sum beamforming method: a acoustic map; b slice of the map for elevation 1, deg The level of the unwanted penetration of acoustic signals is about dB higher in comparison with Christensen beamforming method (Fig 11.8); the level is about dB higher than such obtained using the modified Capon-based beamforming method (Fig 11.6); the level exceeds the level obtained using the heuristic method (Fig 11.7) by more than dB 11.4.5 The Acoustic Camera Responses for Two Point-like Emitters The acoustic camera was applied to check the results of the generation of the acoustic images of two point-like sources of acoustic noise signals The experiment scenario (Fig 11.5) was re-assembled, thus, the sources range, cross-range separation and angular coordinates are close to described above The acoustic camera demo project enables to select the center frequency 10 kHz and the frequency range from 8.913 to 11.22 kHz We assume that an acoustic noise source (Fig 11.5) spectrum shape is uniform in the frequency range The bandwidth 2.308 kHz is wider comparable to used above bandwidth kHz The “acquisition time” is set to s The B&K software “calculation setup” was defined as “default delay and sum” The software indicates that it includes calculation of cross spectra, principal component decomposition, and transducer electronic data sheet application The generated acoustic image is given in the Fig 11.10 Fig 11.10 The Acoustic Camera built-in software output: the acoustic map screenshot with corresponding color bar, center frequency, and frequency range The experiment was repeated in order to record the Acoustic Camera signals for processing with the modified Capon beamforming method using the above approach The frequency range for the modified Capon beamforming method is widened The frequency range was from 8.912 to 11.22 kHz The observation time 0.25 s is times less than used by the demo project The acoustic map threshold level is −5 dB (Fig 11.11) The level of the hollow between the map peaks is better than −3 dB (Fig 11.11) This depth of hollow enables to resolve these two peaks The shift of angular coordinates of these peaks in Fig 11.11 corresponds to the sources shift affected by the mentioned above re-assembling of the experiment scenario Fig 11.11 Acoustic map generated using the modified Capon beamforming method for comparison with the Acoustic Camera built-in software The experiment with observation time 0.25 s was repeated several times that showed the resolution improvement repeatability as well as the variation of the sidelobe level The sidelobe level of the method (Fig 11.11) varies from −3 to −9.7 dB due to: acoustic noise signal level; the Acoustic Camera phase stability; the equipment placement inside the office room Thus, the modified Capon-based beamforming method delivers better resolution than “default delay and sum” regime of the B&K software The improvement is obtained for times less acquisition time The proposed modified Capon-based beamforming method enables to improve the microphone array resolution 11.5 Conclusions This chapter discussed algorithms for acoustic map generation An algorithm based on modified Capon method is proposed It has been tested using acoustic camera software and hardware manufactured by B&K The comparison analysis shows improved resolution characteristic of the newly proposed method in comparison with classical ones and built-in ones in B&K equipment The angular resolution improvement was obtained for center frequency in the range of 0.1–18 kHz The algorithm improves the resolution of acoustic maps generated in Noise Exposure Monitoring The ability to resolve closely placed sources has been shown experimentally Acknowledgments The research work reported in the chapter was partly supported by the Project AComIn “Advanced Computing for Innovation”, grant 316087, funded by the FP7 Capacity Programme (Research Potential of Convergence Regions) References Billingsley, J.: An acoustic telescope Aeronautical Research Council 35/364 (1974) Billingsley, J., Kinns, R.: The acoustic telescope J Sound Vib 48, 485–510 (1976) [CrossRef] Gerges, S., Fonseca, W.D., Dougherty, R.P.: State of the Art Beamforming Software and Hardware for Applications Paper presented at the 16th International Congress on Sound and Vibration, Krakow, 5–9 July 2009 Greguss, P.: Ultrasonic Imaging: Seeing by Sound: the Principles and Widespread Applications of Image Formation by Sonic, Ultrasonic and Other Mechanical Waves Focal Press Limited, London (1980) Born, M., Wolf, E.: Principles of Optics, 7th edn Cambridge University Press, Cambridge (1999) [CrossRef] Dougherty RP (2002) Beamforming in acoustic testing In: Mueller T (ed) Aeroacoustic Measurements Springer Berlin Heidelberg, New York: 62– 97 Sarradj, E., Schulze, C., Zeibig, A.: Einsatz eines Mikrofonarrays zur Trennung von Quellmechanismen, Vortrag auf der Tagung der Deutschen Arbeitsgemeinschaft für Akustik—DAGA, Technischen Universität München, München, 14–17 Marz 2005 Capon, J.: High-resolution frequency-wavenumber spectrum analysis Proc IEEE 57(8), 1408–1418 (1969) [CrossRef] Marple, S.L.: Digital Spectral Analysis with Applications Prentice-Hall Inc., Upper Saddle River (1987) 10 Munier, J., Delisle, G.Y.: Spatial analysis in passive listening using adaptive techniques Proc IEEE 75(11), 1458–1471 (1987) [CrossRef] 11 Ermolayev, V.T., Flaksman, A.G.: Metodi Otsenivaniya Parametrov Istochnikov Signalov i Pomeh, Prinimayemikh Antennoi Reschetkoi (Methods for Estimation of Parameters of Sources of Signals and Interferences Received by Antenna Array) Lobachevsky State University, Nizhny Novgorod (2007) 12 Stoica, P., Moses, R.: Introduction to Spectral Analysis Prentice Hall, Upper Saddle River (1997) [MATH] 13 Lukin, K., Kudriashov, V.V., Vyplavin, P., Palamarchuk, V.: Coherent imaging in the range-azimuth plane using a bistatic radiometer based on antennas with beam synthesizing IEEE Aerosp Electron Syst Mag 29(7), 16–22 (2014) [CrossRef] 14 Lopez, E.S.: Sensor Array Optimization for Multiple Harmonic Sound Source Separation and DOA MA Dissertation, Lund University, Lund (2013) 15 Christensen, J.J., Hald, J.: Beamforming Brüel & Kjær Technical Review 1, 1–48 (2004) 16 Brüel & Kjær Sound and Vibration Measurement A/S: Product Data: PULSE Array-based Noise Source Identification Solutions: Beamforming—Type 8608, Acoustic Holography—Type 8607, Spherical Beamforming—Type 8606 Brüel & Kjær Sound and Vibration Measurement A/S, Nærum (2009) 17 IEEE Instrumentation and Measurement Society: 1588-2002—IEEE Standard for a Precision Clock Synchronization Protocol for Networked Measurement and Control Systems IEEE, USA (2002) 18 Lukin, K., Vyplavin, P.: Integrated sidelobe ratio in noise radar receiver Paper presented at the IEEE 13th International Radar Symposium (IRS), Warsaw, 23–25 May 2012 ... Roumen Kountchev and Kazumi Nakamatsu New Approaches in Intelligent Image Analysis Techniques, Methodologies and Applications Editors Roumen Kountchev Department of Radio Communications and Video Technologies,... Kountchev and Kazumi Nakamatsu (eds.), New Approaches in Intelligent Image Analysis, Intelligent Systems Reference Library 108, DOI 10.1007/978-3-319-32192-9_1 New Approaches for Hierarchical Image. .. the authors in the corresponding scientific area The object of the investigation is new methods, algorithms, and models, aimed at the intelligent analysis of signals and images—single and sequences