Computationally Efficient Scaled Clustering for Perceptually Invisible Image Intensity Tuning

Abstract

Quality and information content of an image is a vital aspect affecting the performance of any artificial intelligence-related task. Improving the quality of an image while maintaining its color and naturalness in a very small time frame is a challenge. In this article, a computationally efficient scaled clustering-based contrast enhancement method is proposed for enhancing perceptually invisible images and preserving color and naturalness. Pixels in the image are grouped in clusters using the fuzzy c-means algorithm and are assigned membership to different clusters. Membership values of the pixels are modified and scaled up using the cluster centers. Based on these modified and scaled cluster memberships, the intensity level of the pixels is modified. The membership-based proportional modification of the intensity values in the scaled domain improves the color and information content of an image. It is scaled down to the original range of intensity levels, using linear scaling to obtain an enhanced image with better contrast and color information. The generated image is free from artifacts and is natural in appearance. The proposed method is simulated and tested on frequently used standard datasets. The simulation results and the average computation time of the proposed method are compared to some of the well-established conventional and latest techniques.