Image Enhancement Using Patch-Based Principal Energy Analysis

The visual quality of a captured image is often degraded by complicated lighting conditions in various real-world environments. This quality deterioration probably leads to the significant performance drop in many algorithms of computer vision, which require high-visibility inputs for precise results. In this paper, a novel method for image enhancement is proposed with the principal energy analysis. Specifically, based on the key observation that the illumination component is dominant over a small local region, the corresponding energy is efficiently separated from the scene reflectance by exploiting the subspace analysis. Owing to this clear separation, the illumination component can be easily adjusted independent of the reflectance layer for better visual aesthetics. In contrast to previous methods that still suffer from the exaggerated or conservative restoration yielding the loss of details and defects of halo artifacts, the proposed scheme has a good ability to enhance the image contrast while successfully preserving the color attribute of the original scene. Moreover, the proposed method is conceptually simple and easy to implement. Experimental results demonstrate the effectiveness of the proposed method even under diverse lighting conditions, e.g., low light, casting shadow, uneven illuminations, and so on, and the superiority of the proposed method over previous approaches introduced in the literature.