What is the result of combining equalized image with a sobel filter?

Hi,

I was trying various things to make text stand out more in images. A hack which I came up with was by applying the sobel filter for vertical then horizontal to the rgb image then adding it an equalized version of the original image (code snippet below). However I couldn't segment or binarize the image, but it seems to make all text and well defined features darker almost as if it added some texture to the image. Please let me know if anyone knows what's going on here, im quite interested to find out whats happening.

IM=imread('image.jpg');

% Apply horizontal and vertical sobel filters:
h=fspecial('sobel');
filt=imfilter(IM,h);
filt=imfilter(filt,h');

% Equalize the image:
r=squeeze(IM(:,:,1));
g=squeeze(IM(:,:,2));
b=squeeze(IM(:,:,3));
rgb=r+g+b;
R=r./rgb;G=g./rgb;B=b./rgb;
EQ=cat(3,R,G,B);

% Resultant image seems to have amplified curves:
IMNEW= filt + EQ;

Answer 1

[Image Analyst]

No, you're doing it all wrong. I started to try to fix your code, but it's better to just start from scratch and do it the right way from step 1. Here, try this way, using imgradient():

    clc;    % Clear the command window.
    close all;  % Close all figures (except those of imtool.)
    imtool close all;  % Close all imtool figures.
    clear;  % Erase all existing variables.
    workspace;  % Make sure the workspace panel is showing.
    format longg;
    format compact;
    fontSize = 20;
    % Read in RGB image.
    rgbImage=imread('onion.png');
    % Display image.
    subplot(3,4, 1);
    imshow(rgbImage);
    title('Original Color Image', 'FontSize', fontSize);
    % Enlarge figure to full screen.
    set(gcf, 'units','normalized','outerposition',[0 0 1 1]);
    % Extract the individual red, green, and blue color channels.
    redChannel = rgbImage(:, :, 1);
    greenChannel = rgbImage(:, :, 2);
    blueChannel = rgbImage(:, :, 3);
    % Display images.
    subplot(3,4, 2);
    imshow(redChannel);
    title('Red Image', 'FontSize', fontSize);
    subplot(3,4, 3);
    imshow(greenChannel);
    title('Green Image', 'FontSize', fontSize);
    subplot(3,4, 4);
    imshow(blueChannel);
    title('Blue Image', 'FontSize', fontSize);

    % Apply Sobel filters:
    sobelFilteredImageR = imgradient(redChannel);
    sobelFilteredImageG = imgradient(greenChannel);
    sobelFilteredImageB = imgradient(blueChannel);
    % Combine them all so we can add to original
    colorSobel = cat(3, sobelFilteredImageR, sobelFilteredImageG, sobelFilteredImageB);
    % Display images.
    subplot(3,4, 6);
    imshow(sobelFilteredImageR, []);
    title('Red Sobel Image', 'FontSize', fontSize);
    subplot(3,4, 7);
    imshow(sobelFilteredImageG, []);
    title('Green Sobel Image', 'FontSize', fontSize);
    subplot(3,4, 8);
    imshow(sobelFilteredImageB, []);
    title('Blue Sobel Image', 'FontSize', fontSize);

    % Add together with three different weightings.
    weighting = 0.1;
    outputImage = rgbImage + uint8(weighting * colorSobel);
    subplot(3,4, 9);
    imshow(outputImage);
    caption = sprintf('Weighting of %.2f', weighting);
    title(caption, 'FontSize', fontSize);

    weighting = 0.4;
    outputImage = rgbImage + uint8(weighting * colorSobel);
    subplot(3,4, 10);
    imshow(outputImage);
    caption = sprintf('Weighting of %.2f', weighting);
    title(caption, 'FontSize', fontSize);

    weighting = 0.7;
    outputImage = rgbImage + uint8(weighting * colorSobel);
    subplot(3,4, 11);
    imshow(outputImage);
    caption = sprintf('Weighting of %.2f', weighting);
    title(caption, 'FontSize', fontSize);

    weighting = 1.0;
    outputImage = rgbImage + uint8(weighting * colorSobel);
    subplot(3,4, 12);
    imshow(outputImage);
    caption = sprintf('Weighting of %.2f', weighting);
    title(caption, 'FontSize', fontSize);

[RGR]

Thanks for the code, but what i'm really after is segmenting speed signs from an image. Problem being you have changing shape, luminosity, and color (dependent on reflections). That's why I use the sobel filters to get a combination or the vertical and horizontal gradients which essentially for edge detection (although I agree it should be applied at the channels individually) the equalized image was to find the most chromatic portions of the image (gray will be gray and acrhomatic values). Speed signs are inherently both chromatic and achromatic depending on what they are reflecting. Thus I wanted to combined the horzintal and vertical sobel filters to rgb images in order to amplify its features (its a rectangle) and then segment that image. I have no clue as to how I would go about segmenting this image to binary. Its a weird problem :S

[Image Analyst]

Sorry I thought that's what you wanted, even though it seemed like a strange thing to do. If what you're really after is to locate road signs, you need to go here http://iris.usc.edu/Vision-Notes/bibliography/active693.html#Road%20Signs,%20Traffic%20Signs,%20Objects%20along%20the%20Road,%20Inspections to find an algorithm that works.

[RGR]

Thanks for the links but I have already come across most of these, the major problem is that the sign is achromatic in a large portion of the scenarios. I actually am doing my final year design project on this. I will upload as soon as its completed, however in order to segment out the road sign the best way is to use NTSC (YIQ) locally thresholded (OTSU method) binary image, which then can be morphologically operated upon then open the image for rectangles (I used a GA to determine the size of the SE). This is a very complex problem as there is no good segmentation technique which can be employed, I checked some of the papers in the link provided above, but most of the papers/Algorithms deal with chromatic sign detection. Though I still haven't been able to browse through all the algorithms for this specific algorithm, I believe that this is too far off topic from the OP and will create a new submission file with my code and post the link here, thanks for the links though :).