BFAGC: A Bias-Free Adaptive Gamma Correction Method for Image Enhancement Nazifa Akter, Sharmin Sultana, Rahat Hossain Faisal and Md. Mostafijur Rahman Dept. of Computer Science and Engineering, University of Barishal, Barishal, Bangladesh sharmin.cse1.bu@gmail.com, nazifa.cse1.bu@gmail.com, rhfaisal@gmail.com and mostafij.csebu@gmail.com Abstract—One of the most interesting and visually appealing areas of image processing is image enhancement. In order to enhance the images, various methods have already been introduced. But, these existing enhancement methods are hardly able to enhance all types of images. To address this issue, Adaptive Gamma Correction (AGC) method has been introduced which classifies images into several classes based on the statistical information of the image and then applies an adaptive gamma correction method. However, AGC method can be affected by outlier pixels and fails to appropriately enhance the images. To solve this problem, we propose an improved version of Adaptive Gamma Correction (AGC) namely Bias-Free Adaptive Gamma Correction (BFAGC). The proposed BFAGC is rarely affected by the outlier pixels. The extensive experimental analysis has been performed to evaluate the efficacy of the proposed BFAGC method. Both the qualitative and quantitative evaluation metrics have shown that BFAGC produces comparatively better enhanced images than AGC and other existing state-of-the-art methods. Keywords—Contrast Enhancement, Image Classification, Im- age Enhancement, Adaptive, Gamma Correction. I. I NTRODUCTION Image enhancement is one of the well known issues in image processing field. Since, most of the images are affected by cloud, noise, low quality of the devices that capture images and the lack of operator expertise, image enhancement is needed to increase the visual quality and to remove unwanted noise of those images [1]. Besides this, image enhancement is necessary because of its expansive range of application in areas such as forensics [2], atmospheric sciences [3], astrophotogra- phy [4], oceanography [1], satellite image analysis [1], medical image processing [5], radar imaging [6], texture analysis and synthesis [1], remote sensing [7], person identification [8], face image analysis [9], machine learning [10], [11], [12], digital photography [13], surveillance and video processing applications [14]. In order to enhance the quality of an image, various contrast enhancement methods have been proposed. These methods involve Histogram Equalization (HE) [15], Brightness Preserving Bi-Histogram Equalization (BBHE) [16], Dualistic Sub-Image Histogram Equalization (DSIHE) [17], Recursively Separated and Weighted Histogram Equalization (RSWHE) [18], Contextual and Variational Contrast (CVC), Layered Difference Representation (LDR) [19], Gamma Correction with Weighting Distribution (AGCWD) [20] and Adaptive Gamma Correction (AGC) [21]. These methods are briefly described in the followings. HE is one of the most fundamental image enhancement techniques because of its efficacy to give better contrast for all types of images [22]. It is a method that expanses the contrast by redistributing the gray level values equally [23]. Although, HE improves the contrast and brightness of an image, it does not always give favorable result. It produces a significant amount of artifacts due to the mean-shift problem [23]. To reduce this problem, BBHE [16] and DSIHE [17] methods are introduced. BBHE [16] method solves the over-enhancement problem by dividing the input image into two sub-images based on the mean of the image. Then, HE is separately applied on both sides of the images. Afterward, two parts are merged to get the final output. For symmetric distribution of input image around its mean, BBHE provides better result. But it may fail for non-symmetric distribution [23]. Considering this issue, DSIHE [17] method is introduced which divides the histogram into two sub-images on the basis of median [17]. Then, each sub-image is equalized like BBHE. Though, DSIHE method neglects the significant mean shift problem in some cases, it does not able to preserve mean brightness. It fails to improve some extents of image or to create artifacts on image [23]. Then, another method namely RSWHE [18] is introduced by combining BBHE [16] and DSIHE [17] meth- ods, which improves the contrast and conserves the brightness of an image. It partitions the histogram into two parts and applies weights. At last, it performs histogram equalization on each part of the weighted histograms. This method has the same time complexity but expands DSIHE by including multi-equalization to reduce the generation of artifacts. Except these techniques, some other techniques have al- ready been introduced. Among these, CVC [24] method uses inter pixel contextual information. This information is taken, and then enhancement is performed on its neighboring pixels which creates a 2D histogram [24]. For creating a better visual quality output image, this method is an effective one. But, the computational complexity of this method becomes higher. Another method, LDR [19] divides an image into multiple layers. Then, it derives a transformation function for each layer [19]. After performing those transformation functions, it merges all the layers to get final output. Another method, combining traditional gamma correction and histogram equal- ization methods are AGCWD [20]. This method first analyzes the histogram of the image then weighting distribution is used and then the Gamma Correction method is applied to the image to get the enhanced output image [20]. Although,