Which techniques can be applied instead of histogram binning?

Hi I read a paper using latent dirichlet allocation (LDA), which author has created the intensity intervals for images to create the words ( intensity intervals), and the number pixels that fall into each interval (bin) is counted.However, this leads to poor performance in detecting low-contrast pixel intensities. My question is which technique can be used instead of histogram binning, how we can address this issue? I really do not know how we can tackle with this?

Your help is appreciated :)

Many thanks

[Jan]

Please define "poor performance" with any details, because we cannot guess them. The term "faint" sounds to imply, that the results have a "poor performance".

[Sara Salimi]

Thanks, I revised the question :)

[Jan]

Does "performance" concern the computing time or are the results "poor", because the pixels with low intensitied cannot be distinguished efficiently? You did not mention, what you want to achieve. All you mention is, that a certain method is "poor". But how can this be used to suggest a "better" method?

[Image Analyst]

The low contrast pixels are there (not missing from the image), so what do you mean by "detecting low-contrast pixel intensities"? Contrast as computed over what region or window size? Why is a histogram needed? Why not just take the Laplacian of the image, with conv2(), and find values close to zero?

[Sara Salimi]

Dear all,

Apologies for poor English, so I attached the paper, as you can see in the page 433 (first paragraph),

1. they have mentioned the problem with histogram binning, which results vary based on the number of bins in the histogram. Which technique can be used for creating words for LDA instead of histogram binning?
2. in the forth paragraph, they have mentioned problem of computation for calculating Gibbs sampling for each pixel. How can we solve this? by calculating that for each region? how?

I will be thankful if somebody can help me. I really need help. Or if you know any reference that I can get any idea for this, I will be thankful if you introduce it to me.

Your help is appreciated

regards