Novel Unsupervised Learning Architecture for Exposure-Based Classification and Enhancement

Numerous imaging applications are affected by the poor quality of images caused by poor illuminating conditions, contrast degradation, and unwanted noise. These effects create noticeable artifacts in an indeterministic selective manner, where some parts of the image are modified, and some parts of the image are uninfluenced. Thus, the classification of an image into various sections and, then, segmentwise application of imaging algorithms are a preferable solution. This article focuses on classifying an image into three categories as under–well–over exposed regions. This article introduces the concept of multilevel superpixel-based classification. Superpixel stores the local integrity and color similarity of an image; hence, an image is initially classified into an experimentally predetermined number of superpixels. Then, a novel algorithm depending upon the superpixel contrast, entropy, and statistical distribution of illumination classifies it into an under–well–over exposed region. Then, with an increased number of superpixels, we reiterate the whole process. The regions classified into the same category in both iterations perform as the training datasets for the support-vector-machine (SVM) classifier. Finally, the trained SVM classifies the ambiguous regions obtained from multilevel superpixel classification. Both qualitative and visual results show the superior performance of the proposed method over the state-of-the-art methods.