what are suitable texture features for these images

dear i have these images and i want to recognition between them based on texture features [density of yellow color]

can any one suggest for me a single value texture feature ??

Answer 1

[Image Analyst]

First of all, you need to use crossed polarizers to get rid of the white specular reflections.

Next, I don't see how "density of yellow" is a texture feature. It's more of a color/intensity/radiometric quantity since it's the wavelength/spectrum of light rather than a spatial variation.

What I'd probably do for a start is to compute two classes: reddish, and yellowish/greenish. Look at the color gamut in HSV color space, looking down along the V axis:

You can see how the yellowish/greenish pixels are separated from the reddish/purplish pixels. I'd calculate the total number of pixels for each color class as well as the area fraction for each color class. I'd also give the mean LAB color of each, and if you have a "true/nominal/reference" color then I'd give the delta E (the "color difference" between the actual object and it's reference color. See my delta E demo in my File Exchange. Also, see my "Color detection by hue" full demo in my File Exchange: http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862

[mangood UK]

dear Image Analyst

your answer amazing, i will never got such answer in any place,

god bless you.

i will read and test more about you answer and back again

best regards

[Image Analyst]

You're welcome. If you do want to measure texture, then take an image, such as the "v" channel of the HSV image, and run entropyfilt() or stdfilt() on it. Areas with high variation will be bright and smooth areas will be dark. You can segment areas based on their texture and give the mean texture in those areas. For example you could use color segmentation to find green and red areas, and then combine that with thresholded texture images to find textured green, smooth green, textured red, and smooth red regions.

[mangood UK]

 welcome too.
about this Sentence
""You can segment areas based on their texture and give the mean texture in those areas""
 do you have any  demo??
regards

[Image Analyst]

Something like

hsv = rgb2hsv(rgbImage);
v = hsv(:,:,3);
textureImage = stdfilt(v);
highTexture = textureImage > 3; % or whatever number you pick.
pixelsInHighTextureRegions = v(highTexture);
meanOfPixelsInHighTextureRegions = mean(pixelsInHighTextureRegions);

and so on....

[mangood UK]

ok in matlab code its more clear

final ask [apologies for many ask]

about ""color segmentation to find green and red areas, and then combine that with thresholded texture images to find textured green""

what do you mean by ""combine"" with ""thresholded texture images"" is it add operation !!

[Image Analyst]

You could get two binary images. One where you segmented based on hue or saturation or something and one where you did it based on texture. Then to get pixels that are both red and smooth you AND the images

smoothTexturePixels = textureImage < 3; % or whatever.
redPixels = (h < 0.3 | h > 0.9) & (s > 0.2); % or whatever.
% Now AND
bothRedAndSmooth = redPixels  & smoothTexturePixels;

Similar for other colors and textures.

[mangood UK]

ok dear Image Analyst [we not has your real name], its more clear now

i will start testing and i will report for any issue

best regards

[Image Analyst]

OK. I prefer to remain anonymous, though some people know who I am (like people at the Mathworks and some others). If it works after you test it, then please mark my Answer as "Accepted."

[mangood UK]

as you like Image Analyst

have a nice day

regards