Advanced Histogram Equalization Based on a Hybrid Saliency Map and Novel Visual Prior

The traditional grayscale histogram of an input image is constructed by simply counting its pixels. Hence, the classical histogram equalization (HE) technique has fundamental defects such as overenhancement, underenhancement, and brightness drifting. This paper proposes an advanced HE based on a hybrid saliency map and a novel visual prior to addressing the defects mentioned above. First, the texture saliency map and attention weight map are constructed based on the texture saliency and visual attention mechanism. Later, the hybrid saliency map that is obtained by fusing the texture and attention weight maps is used to derive the saliency histogram. Then, a novel visual prior, the narrow dynamic range prior (NDP), is proposed, and the saliency histogram is modified by calculating the optimal parameter in combination with a binary optimization model. Next, the cumulative distribution function (CDF) is rectified to control the brightness. Finally, the hybrid saliency map is applied again for local enhancement. Compared with several state-of-the-art algorithms qualitatively and quantitatively, the proposed algorithm effectively improves the contrast of the image, generates better subjective visual perception, and presents better performance broadly.