segmentation of image
in [Matlab]
|
Prev: regexprep tokens
Next: svmtrain problem
From: Vihang Patil on 3 Jul 2010 06:41 Hello, I am trying to achieve segmentation of an image, to futher do OCR on it. I am able to remove the background of the image and able to get the 3 letters that I am interested in the image. Now, I want to further segment this letters individualy so that I can do OCR and achieve the letters in text format. But since there is very little gap in between the letters and also since there is very little variation in the grascale, I am unable to do so. Could anyone help me on this one. The code for segmenting the 3 letters in the image is as follows; I = imread('1.bmp'); gray = rgb2gray(I); figure,imshow(I); J = imcrop(gray); %create a small ROI around the letters either FRL or RRL figure,imshow(J); J2 = ipexbatchDetectCells(J); J3 = imsubtract(J2,J); bw = im2bw(J3,graythresh(J3)); bw_fill = imfill(bw,'holes'); % figure,imshow(bw_fill); SE = strel('rectangle',[3 3]); bw_fill1 = imdilate(bw_fill,SE); bw_fill1 = imfill(bw_fill1,'holes'); L = bwlabel(bw_fill1); stats = regionprops(L,'area'); area = [stats.Area]; if length(area) > 1 bw_fill1 = bwareaopen(bw_fill1 ,(max(area) - 5)); end figure,imshow(bw_fill1); K = J; K(~bw_fill1) = 0; figure,imshow(K); The sample files are located here http://drop.io/vihdrop Thanks Vihang
From: ImageAnalyst on 3 Jul 2010 09:11 On Jul 3, 6:41 am, "Vihang Patil" <vihang_pa...(a)yahoo.com> wrote: > [snip] since there is very little gap in between the letters and also since there is very little variation in the grascale, I am unable to do so. [snip] > Vihang -------------------------------------------------------------------------- It looks like you have good control over the lighting and camera. As most people know it's better to start with a good image rather than a bad one that you try to "fix up" later with image processing. Therefore to improve the problem issues you listed, I recommend that you get a camera with a higher resolution and zoom in some more so that your pair of letter strings occupy more pixels. It looks like you're designing a machine vision system of your own rather than using one of the numerous commercial systems for which this situation is cheaply and easily handled. The system consists of two parts: the image acquisition and the image analysis. You're working on the image analysis part now but I urge you not to just assume that the work you've done on the image capture part is perfected and finalized. Improve the image capture part and your image analysis part will be easier. If you're stuck with the images because the vendor did only the image capture part, then ask them to finish the job and do it right to deliver a complete turnkey system (any decent vendor will be able to do that). If you're stuck with the images because this is not a real machine vision project but just some student exercise, then let us know that too because there are some things that can be done if you can't do it right (by improving the image capture). -ImageAnalyst
From: Vihang Patil on 3 Jul 2010 10:14 ImageAnalyst <imageanalyst(a)mailinator.com> wrote in message <759cc292-fab0-4385-b254-360363d8dd14(a)i31g2000yqm.googlegroups.com>... If you're stuck with the images because this is not a real > machine vision project but just some student exercise, then let us > know that too because there are some things that can be done if you > can't do it right (by improving the image capture). > -ImageAnalyst Thanks ImageAnalyst for the suggestion, but as of now, I am stuck with these images only. Yes, its a machine vision application, but I will have to manage with these images only. Any suggestion on the image processing part? Vihang
From: ImageAnalyst on 3 Jul 2010 10:33 Vihang Patil: I don't have your function J2 = ipexbatchDetectCells(J); Can you also post the intermediate images that you've obtained via your code?
From: Vihang Patil on 3 Jul 2010 11:01
ImageAnalyst <imageanalyst(a)mailinator.com> wrote in message <cd82e80e-2c98-405c-9f8d-44ea2c55832a(a)d16g2000yqb.googlegroups.com>... > Vihang Patil: > I don't have your function > J2 = ipexbatchDetectCells(J); > Can you also post the intermediate images that you've obtained via > your code? Dear ImageAnalyst ipexbatchDetectCells is an built in function of the Image Processing Toolbox in R2006b. Nevertheless, I am posting it here function segmentedCells = ipexbatchDetectCells(I) %ipexbatchDetectCells Algorithm to detect cells in image. % segmentedCells = ipexbatchDetectCells(I) detects cells in the cell % image I and returns the result in segmentedCells. % % Supports batch processing demo, ipexbatch.m, % ("Batch Processing of Image Files Using Distributed Computing"). % Copyright 2005 The MathWorks, Inc. % $Revision: 1.1.6.1 $ $Date: 2005/06/20 03:09:37 $ % Use |edge| and the Sobel operator to calculate the threshold % value. Tune the threshold value and use |edge| again to obtain a % binary mask that contains the segmented cell. [junk threshold] = edge(I, 'sobel'); fudgeFactor = .5; BW = edge(I,'sobel', threshold * fudgeFactor); se90 = strel('line', 3, 90); se0 = strel('line', 3, 0); BWdilate = imdilate(BW, [se90 se0]); BWnobord = imclearborder(BWdilate, 4); BWopen = bwareaopen(BWnobord,200); BWclose = bwmorph(BWopen,'close'); BWfill = imfill(BWclose, 'holes'); BWoutline = bwperim(BWfill); segmentedCells = I; segmentedCells(BWoutline) = 255; Vihang |