Source-linked AI summary
A Comparative Study of Histogram Equalization Based Image Enhancement Techniques for Brightness Preservation and Contrast Enhancement
Omprakash Patel, Yogendra P. S. Maravi, Sanjeev Sharma
TL;DR
The paper addresses the difficulty of enhancing contrast without substantially changing image brightness. It reviews histogram-equalization extensions, including mean-, median-, and recursively partitioned methods, and compares them using brightness, quality, structural, and entropy measures. The study reports that RSWHE-M provides better contrast enhancement and structural similarity, while RSWHE-D offers the best brightness preservation among the reviewed methods.
Problem
Standard histogram equalization can produce output mean brightness significantly different from the input, limiting its use where brightness preservation matters.
Method
The paper reviews histogram-equalization extensions that split histograms using input means, medians, recursive rules, or optimized thresholds before independently equalizing sub-images.
Results
RSWHE-M provides better contrast enhancement and structural similarity, while RSWHE-D is the best method for brightness preservation in the study’s comparison.
Takeaways & Limitations
Recursive and weighted histogram-equalization variants offer brightness preservation alongside contrast enhancement, with different variants favored for different evaluation criteria.
Abstract
from arXiv · showhide
Histogram Equalization is a contrast enhancement technique in the image processing which uses the histogram of image. However histogram equalization is not the best method for contrast enhancement because the mean brightness of the output image is significantly different from the input image. There are several extensions of histogram equalization has been proposed to overcome the brightness preservation challenge. Contrast enhancement using brightness preserving bi-histogram equalization (BBHE) and Dualistic sub image histogram equalization (DSIHE) which divides the image histogram into two parts based on the input mean and median respectively then equalizes each sub histogram independently. This paper provides review of different popular histogram equalization techniques and experimental study based on the absolute mean brightness error (AMBE), peak signal to noise ratio (PSNR), Structure similarity index (SSI) and Entropy.
1. INTRODUCTION
The paper reviews histogram-equalization methods for contrast enhancement while focusing on the persistent challenge of preserving image brightness. It introduces extensions that partition the histogram using mean, median, recursive, or optimized thresholds.
- Motivation: Histogram equalization spreads intensities across the available range but can shift mean brightness toward the middle gray level.This mean-shift makes standard HE unsuitable when brightness preservation is required.
- Brightness-preserving extensions: BBHE divides the histogram at the input mean and independently equalizes the lower and upper sub-histograms within corresponding intensity ranges.The lower range spans the minimum gray level to the mean, while the upper range spans the mean to the maximum.
- Brightness-preserving extensions: DSIHE instead partitions the histogram at cumulative probability 0.5, producing two equal-area sub-images before independent equalization.The reviewed work reports DSIHE as better than BBHE for brightness preservation and entropy.
- Recursive methods: RMSHE and RSIHE recursively repeat mean- or median-based separation, bringing output brightness closer to the input and enabling scalable preservation.RSIHE uses cumulative distribution 0.5, whereas RMSHE uses respective sub-histogram means.
- Further extensions: MMBEBHE selects a threshold that minimizes absolute mean brightness error, while RSWHE recursively weights sub-histograms before equalization.RSWHE applies a normalized power-law weighting process to its recursively formed sub-histograms.
B. Brightness Preserving Bi-Histogram Equalization (BBHE)-
BBHE partitions an input histogram at its mean and equalizes the resulting sub-images independently over separate intensity ranges.
- BBHE decomposition: BBHE decomposes the input image into lower and upper sub-images using the input mean as the separation point.The lower sub-image contains levels up to the mean, and the upper sub-image contains higher levels.
- BBHE transformation: The two BBHE sub-images are equalized independently and then composed to form the output image.Each transformation uses the intensity range assigned to its corresponding sub-image.
C. Equal Area Dualistic Sub-Image Histogram Equalization (DSIHE):-
DSIHE extends bi-histogram equalization by splitting the image at the median gray level defined by cumulative probability 0.5, then equalizing both sub-images.
- DSIHE decomposition: DSIHE separates the input histogram at the gray level whose cumulative distribution equals 0.5 rather than at the input mean.This produces dark and bright sub-images with equal pixel counts.
- DSIHE transformation: The method targets maximization of Shannon entropy by independently histogram-equalizing the two separated sub-images and recomposing them.The lower and upper sub-images are mapped to their respective intensity ranges before composition.
D. Recursive Mean Separate Histogram Equalization (RMSHE):-
RMSHE generalizes BBHE by recursively separating each newly formed histogram according to its respective mean, providing scalable brightness preservation.
- RMSHE: RMSHE repeatedly splits each sub-histogram at its respective mean instead of performing BBHE’s single separation.Increasing recursion levels brings output mean brightness closer to the input mean brightness.
E. Minimum Mean Brightness Error Bi-Histogram Equalization in Contrast Enhancement (MMBEBHE):-
MMBEBHE extends BBHE by selecting a histogram threshold that minimizes absolute mean brightness error before separately equalizing the resulting sub-histograms.
- MMBEBHE separates the input histogram at a threshold selected to minimize absolute mean brightness error, then equalizes both sub-histograms as in BBHE.It evaluates AMBE across threshold levels, chooses the minimum-error threshold XT, and applies histogram equalization to the separated histograms.
F. Recursive Sub-Image Histogram Equalization Applied To Gray Scale Image (RSIHE):-
RSIHE generalizes DSIHE by recursively splitting the histogram at cumulative probability 0.5, producing equal-pixel sub-images for scalable brightness preservation.
- RSIHE recursively divides the input histogram at cumulative probability 0.5, creating sub-images with equal pixel counts and scalable brightness preservation.The method applies the equal-probability split for a specified number of recursion levels.
- The segmentation module can recursively split a histogram by either the mean or median of its sub-histograms.This module is described as operating up to a specified recursion level.
- At recursion level r, histogram segmentation generates 2^r sub-histograms.
- The weighting module modifies each sub-histogram probability density function using a normalized power-law function.The weighting process uses the original probability density and cumulative probability values for each sub-histogram.
- The weighting parameter b controls the balance between output mean brightness and contrast enhancement, with satisfactory quality reported when b is near the stated experimental value.After weighting, the histogram is normalized before subsequent processing.
3. Histogram equalization Module:-
The histogram equalization module independently equalizes every weighted, normalized sub-histogram and combines the resulting sub-images into the final RSWHE output.
- Each of the 2^r weighted, normalized sub-histograms is equalized separately, and the resulting sub-images are finally combined.The sub-histograms are equalized over their respective intensity ranges before reconstruction of the output image.
2. EXPERIMENTAL RESULTS
The study evaluates histogram equalization methods using brightness preservation, contrast enhancement, image similarity, entropy, and visual inspection across standard images. RSWHE-M performs best for brightness preservation and PSNR, while RSWHE-D is generally the second-best method.
- Experimental setup: The experiments compare HE, BBHE, DSIHE, RMSHE, RSIHE, MMBEBHE, and RSWHE methods on standard images.The evaluation includes quantitative metrics and visual results for the images Couple and Einstein.
- Evaluation metrics: Brightness preservation is assessed with AMBE, while PSNR, SSIM, and entropy assess image quality and enhancement characteristics.Entropy measures the richness of details, and SSIM measures similarity against a reference image.
- Assessment of Brightness Preservation: Lower brightness error indicates better brightness preservation.The study defines minimum brightness error as better brightness preservation.
- Assessment of Brightness Preservation: RSWHE-M achieves the best brightness preservation, while RSWHE-D ranks second.The comparison is based on the methods' performance in Table-1.
- Assessment of contrast enhancement: RSWHE-M has the highest PSNR for each tested image and the highest average PSNR among the compared methods.RSWHE-D also performs better than the other histogram equalization methods.
- Inspection of visual quality: The visual inspection includes processed versions of the Couple and Einstein images produced by the tested methods.Figures 4 and 5 present the corresponding method outputs.
3. CONCLUSION
The study compares histogram enhancement methods and reports that RMSHE, DSIHE, RSIHE, and RSWHE improve brightness preservation over HE, BBHE, and DSIHE, while RSWHE-M provides strong contrast and structural similarity.
- The comparative study evaluates histogram enhancement methods for brightness preservation and contrast enhancement.
- Brightness preservation is handled poorly by HE, BBHE, and DSIHE but better by RMSHE, RSIHE, and RSWHE.
- RSWHE-M offers better contrast enhancement and structure similarity, while RSWHE provides the second-best brightness preservation.
- RSWHE-M combines improved enhancement and structural similarity with the study's broader brightness-preservation comparison.
- The paper notes that substantial work has already been conducted in this field, while identifying space for future work.
AUTHORS
The authors are affiliated with the School of Information Technology at UTD, RGPV, Bhopal, and work across image enhancement, medical image processing, and mobile computing.
- The authors are affiliated with the School of Information Technology, UTD, RGPV, Bhopal, as assistant or associate professors.
- One author's research interests include image enhancement and medical image processing.
- Another author's research interest is mobile computing, with memberships in CSI, ACM, and CSTA.