Histogram equalization

I have added histogram equalization to the Image Algorithms demo. This method scans through an image, inspecting each pixel and creates a histogram. If a pixel p_q of grayscale intensity g_q is found at location (x,y), the Histogram H is incremented at position g_q for all grayscale levels L = 0 .. G-1, where G = 256 for grayscale images. A cumulative histogram is then calculated and the resulting pixel value is the ratio between the cumulative histogram and the total amounts of pixels in the image multiplied with G - 1.

More formally.

ALGORITHM - Histogram equalization.

1. For an NxM image of G gray levels, create an array H (histogram) of length G initialized with default intensity value 0.

2. Form the image histogram. Scan every pixel and increment relevant position in the histogram H. If pixel p has intensity g_p, perform:
H[g_p] = H[g_p + 1.

3. Form the cumulative image histogram H_c. Set:
H_c[0] = H[0]
H_c[p] = H_c[p-1] + H[p] for all pixel intensities p where p = 0,1,2... G-1

4. Rescan the image and calculate the corrected intensity, which it output.
Let T[p] = round(((G-1)/(N*M)) * H_c[p]. For a pixel of intensity p in the image
at position x,y, use T[p] to find the correct adjusted value.

5. Display image results.

END.


Note about step 4. Use the (double) cast when doing the math.. In general, histogram equalization is all about creating a Histogram of an image, then form a cumulative histogram, and then use the ratio between the total amount of pixels N*M and multiply with the cumulative histogram H_c[p] and then multiply with (G-1), usually 255.. This will set the most luminous pixels in the image near full intensity, will the less luminous pixels towards zero intensity, increasing the dynamic range and contrast in the image.


Note about the demo, I have added also Hough transform (tested some basic images) and
filters for displaying histogram of the histogram equalization. Test them out. Hough transform shown the sinusoidal images in the parameter space and should be considered as not finished code.

A demonstration follows.



This is equalized to:



Download the demo here:
Image algorithms demo [2,7 Mb zip]

To summarize, to understand histogram equalization is to consider the cumulative histogram and scale this with the ratio with the total amount of pixels multiplied with 255 (G-1).. The result is that the brightest pixels are moved towards full intensity 255 and the darkest pixels are moved towards 0, increasing the contras and dynamic range.

Share this article on LinkedIn.