Machine learning hyperparameter selection for Contrast Limited Adaptive Histogram Equalization

Abstract

Contrast enhancement algorithms have been evolved through last decades to meet the requirement of its objectives. Actually, there are two main objectives while enhancing the contrast of an image: (i) improve its appearance for visual interpretation and (ii) facilitate/increase the performance of subsequent tasks (e.g., image analysis, object detection, and image segmentation). Most of the contrast enhancement techniques are based on histogram modifications, which can be performed globally or locally. The Contrast Limited Adaptive Histogram Equalization (CLAHE) is a method which can overcome the limitations of global approaches by performing local contrast enhancement. However, this method relies on two essential hyperparameters: the number of tiles and the clip limit. An improper hyperparameter selection may heavily decrease the image quality toward its degradation. Considering the lack of methods to efficiently determine these hyperparameters, this article presents a learning-based hyperparameter selection method for the CLAHE technique. The proposed supervised method was built and evaluated using contrast distortions from well-known image quality assessment datasets. Also, we introduce a more challenging dataset containing over 6200 images with a large range of contrast and intensity variations. The results show the efficiency of the proposed approach in predicting CLAHE hyperparameters with up to 0.014 RMSE and 0.935 R 2 values. Also, our method overcomes both experimented baselines by enhancing image contrast while keeping its natural aspect.

Multidimensional Contrast Limited Adaptive Histogram Equalization

“…In the original formulation of CLAHE in 2D [3; 4], the local histogram ranges for all kernels are the same, which works well when the kernels contain intensities in a similarly wide range. Tuning of the trade-off between noise amplification and the signal enhancement is then achieved through selection of the kernel size and the clip limit [30]. However, if different patches of the image data contain local features within vastly different but narrow intensity ranges, they may accumulate in very few histogram bins with values specified globally.…”

“…For both the application of 3D and 4D CLAHE, we used a clip limit of 0.02 and local histograms with 256 greylevel bins. The kernel size for 4D CLAHE was (30,30,15,20) and for 3D CLAHE the same kernel size for the first three dimensions, or (k x , k y , E), was used. Both GHR and AHR settings were tested for comparison.…”

Nipple Localization in Automated Whole Breast Ultrasound Coronal Scans Using Ensemble Learning

“…It employs a local transformation function that enhances each pixel derived from a neighborhood region by applying the histogram equalization on non-overlapping regions of the images by means of interpolation that is used to correct the variations among the borders. 24 It has two parameters, namely the number of tiles and the clip limit. The clip limit controls the over amplification of noise and tile measure size of the non-overlapping region.…”

Detection of Covid-19 in Chest X-ray Image using CLAHE and Convolutional Neural Network

“…7. The key factors in CLAHE are clip limit (CL) and the number of tiles (NT) [19]. The clip limit is a system regulating noise amplification so that the height of the clip does not exceed the specified limit.…”