Image denoising techniques to improve the performance on optical character recognition.
!"# $%&$ ' ()*+++ , *' !' /!0-"1 2*'( *+'1,*3 1 2 4 !"#$ "% #%& ' " 3 (! )!** )+,-+.,+,/0#1 2*2 )!+,-+.,+,/0#1 * )3!* )2!*'" )4!5* )6!5* )7!898989 &$:89 " );!5*$ )!5* <:% #1 )!*<: #12 )!*<: #12 4 (= =!5*" =!5* 5 I. (*, 1. ++, %+ %%,*" % &% > &? ! %**>+', & ? @: & % % 2. /+++*' - &! "?& ? - ! % - !-# < II. 4564 * 0"%% %A: 6 % " B% % % * ! % % % >&% & % 5 / *C *!-%%+-=B, % ! %: % @ " 1. < * *# 2. % @& "*%&@ D%:@ & 7 3. & A 4. & B%& % < E" 1. E" 2. =" 3. '" 4. F"" E# 5. 6. ' 7. #$ 8. #<: III. 4* a. 47 i. 17 E""0* $+,0:& !G& *-*%" 8 * /# 3 3 2 ; 6 ; 4 3 7 3 ; 6 ; )** D*%* %# H% ! ; 6 4 3 7 ; %*;% ** )! $08989 8989 9 I!1+"J, K!1+"J, *II KK 8"9 8"9 * 8*989!108*L* I98LK9 8989 $08*989!18 "J98"J9 ! - &?% # <! )+,-+.,+,/0 #12*2 10 [...]... detail of the input document, we will have a chart expressing greyscale In this chart, there are some top expressing the objects of the image The documentary image has 2 top The first top will express characters, the other will express the background The best threshold is the threshold that the difference between two areas of the historam graph is maximum The value σ b2 is defined : Ratio of the area... algorithm: it is one of the oldest and the most used binarization techniques in image processing Assuming a bimodal distribution within the gray scale histograms of the image, it aims automatically at selecting an optimal threshold T to minimize the within-class variance of two modes, or equiv-alently to maximize their between-class variance The details of the algorithm: After the statistic about the greyscale... probability over the area j Is the probability of gray level i, and the sum of all leve is one, thus we have: 17 j with the whole image: I −1 ∑ p = 1, i= 0 i I want to show you the historam of the gray image In regularly, there are 256 different values in the document image The mean μ1(t) is caculated by the formula: j = 1,2 Some experiments: [a] [c] [b] [d] Picture 8 [a] The noised image [b ]The image after... noise 4 The test set: Light intensity increase 10,20,30,40% 5 The test set: Light intensity decrease 10,20,30,40% 6 The test set: the random character color 7 The testset: the background color 8 The testset: the random characters and background color The result of module image denoising is evaluated by using the Nom-OCR The evaluation process is: - Recognize the noised image - Recognize the denoided image. .. window) then the mean of DA(x) is calculated and replaced it at the center pixel X(i,j) of the window.SD = 20, SF =2 With the size of scanning window is 3x3 then W = 3 b The research noise reduction algorithms by the threshold methds i Global threshold The basis of the algorithm: Using the only threshold with all of the image pixel of the image, which causes bad consequences for distinguishing characters... suggested an image denoising two-step system The first one, the noise image will pass a binary proccess, then, it will become a binary image The next one, it will be denoised by using filters After the research and experiments, we decided to use local adaptive threshold thanks to Otsu algorithm at the first step and at the second step, we will use median filter to denoise 21 Build the automaticaly test... and the background On the other hand, the algorithm can not recognize characters if the objects or characters of the colour images have more light colour than that of the ordinary document Counting up this value threshold can be based on some formulas Step 1:Initial the initial threshold T for all pixels(regularly, T is the average value of histogram image) Step 2: Using threshold T to separate the image. .. Introduction about the Nom optical characters recognition (written by LES-NOM) And this system will be used to evaluate the image denoising module The main module of the Nom-OCR o Module create the tranining data o Module image preprocessing o Module tranning data o Module crop character and recognition Experiment and result IV 4.1 Build Module Image denoising From two main research directions we introduced,... result With the images having many colour objects, it has some wrong results The caculation threshold consists of some repeats in the second steps The thresholds value is only closed and isn’t completely exact in every objects The algrithm has difficulties with the execution time, it takes too much time to calculate the global threshold Some experiments: 15 Picture 6: Experiment using the global threshold... repectively with their little part and when totalling, we will have a binaried big image - The characteristics of the algorithm: Having good results for image noise by the small areas or image - is noised by the light with multi directions and intensities The result of this algorithm is better than that of the global threshold algorithm iii The global threshold using the Otsu’s method 16 The basis of the algorithm: . L1 0 8*L* I98LK9 P P 0 ++ 0 8*989J;,QR,O $ 0 8*9891 0 8*989 PO $ 0 8*989!1J; P 5*! 12 )2*'" iii !<@ i p 0 % %! 17 ,1 1 0 = ∑ − = I i i p 0 0 %24 T +,! . )+,-+.,+,/ 0 #12*2 10 )+,-+.,+,/ 0 #1* ii.