Acceleration of Histogram-Based Contrast Enhancement via Selective Downsampling

Gang Cao1*, Huawei Tian2, Lifang Yu3, Xianglin Huang1 and Yongbin Wang1 1School of Computer Science, Communication University of China, Beijing, China 2People’s Public Security University of China, Beijing, China 3School of Information Engineering, Beijing Institute of Graphic Communication, Beijing, China Correspondence author: gangcao@cuc.edu.cn Abstract In this paper, we propose a general framework to accelerate the universal histogram-based image contrast enhancement (CE) algorithms. Both spatial and gray-level selective down- sampling of digital images are adopted to decrease computational cost, while the visual quality of enhanced images is still preserved and without apparent degradation. Mapping function calibration is novelly proposed to reconstruct the pixel mapping on the gray levels missed by downsampling. As two case studies, accelerations of histogram equalization (HE) and the state-of-the-art global CE algorithm, i.e., spatial mutual information and PageRank (SMIRANK), are presented detailedly. Both quantitative and qualitative assessment results have verified the effectiveness of our proposed CE acceleration framework. In typical tests, computational efficiencies of HE and SMIRANK have been speeded up by about 3.9 and 13.5 times, respectively. Index terms Image processing, contrast enhancement, acceleration, speed up, downsampling, histogram.

1. Introduction Contrast enhancement (CE) of digital images refers to the operations which improve the perceived contrast. Such contrast is typically defined as the dynamic range of pixel gray- levels within global or local image regions. CE is a widely used image enhancement tool in real applications [1]. Generally, a good CE technique is expected to have: 1) more contrast improvement with less image distortion; 2) low computational cost. In consideration of its importance in image processing, plenty of previous works have Acceleration of Histogram-Based Contrast Enhancement via Selective Downsampling 2 presented image CE techniques. In terms of the mapping applied to pixel gray-levels, the existing CE algorithms can be generally categorized as global, local and hybrid ones [2]. Global CE modifies an image via an identical pixel value mapping, such that the gray-level histogram of the processed image resembles the desired one and becomes more spread than that of the original image [2, 3]. Local CE improves contrast by altering pixels in terms of local properties, and typically operates in the image transform domains, such as the discrete cosine transform (DCT) [4] and the discrete wavelet transform (DWT) [5]. Local CE can also be enforced by adaptively applying the global CE to local image regions [3]. Hybrid CE, which combines the global and local CE together, can improve the unified perception of both global and local contrasts [6]. Note that most of existing global CE techniques need to use the gray-level or transform coefficient histogram of input images. As summarized in [3], histogram modification-based CE received the most attention due to straightforward and intuitive implementation qualities. One popular global CE method is histogram equalization (HE) [1], which improves contrast by redistributing the probability density of gray-levels towards uniformity. The prominent merit of HE lies in its high computational efficiency. However, HE might incur excessive enhancement and unnatural artifacts on the images with high peaks in histograms. In order to attenuate such deficiency, lots of improved HE algorithms [3, 7-11] have been proposed, where the histogram modification framework (HMF) [3] is an influential one. HMF treats CE as an optimization problem by minimizing a cost function which includes the penalty of the histogram deviation from original to uniform. Gu et al. [10] proposed an optimal his- togram mapping for automatic CE based on a novel reduced reference image quality metric for contrast change. In [11], a complete HMF is presented by integrating the automatic parameter selection via saliency preservation. T. Celik [6] proposed spatial entropy based CE (SECE) by novelly incorporating the spatial distribution characteristics of pixels into the design of gray-level mapping function. SECE can always slightly improve image contrast without incurring serious image quality degradation. Recently, T. Celik [2] proposes the state-of-the-art global image CE method, SMIRANK, by using spatial mutual information of pixels and PageRank. Although good enhancement quality is achieved, such a algorithm runs rather slower than most of other CE algorithms including HE, HMF, SECE and the adaptive gamma correction with weighting distribution (AGCWD) [12]. Besides, the trans- form coefficient histogram has also been used in CE design [4]. Low computational complexity is an important requirement for the real ap

…(Full text truncated)…