A Region-based Histogram and Fusion Technique for Enhancing Backlit Images for Cell Phone Applications

Abstract

Many cell phone cameras perform poorly in backlighting situations due to low dynamic range, which then leads to the creation of low-quality pictures known as backlit images. Conventional image enhancement algorithms are not well suited to improve the quality of backlit images. Over-saturation or a loss of contrast are typical outcomes when these methods are applied. In this thesis, a novel image enhancement algorithm is presented for improving the visual perception of a single backlit image. The algorithm uses a region-based histogram specification scheme in combination with the discrete wavelet transform image fusion to correct exposure disparities between foreground and background scenes. Computer simulations in MATLAB R2018a and on a dataset of 162 backlit images revealed that the proposed algorithm significantly improves the backlit image's visual perception without distorting colours or adding artefacts. The Peak Signal-to-Noise Ratio, Structural Similarity Index Measure, and Naturalness Image Quality Evaluator metrics objectively validated these results. The algorithm produced PSNR values ranging from 19 dB to 30 dB for images with low backlit degradation while retaining more than (60-70) % structural similarity to the inputs. Lower PSNR and SSIM values were consistent with severely degraded images. These findings agreed with the outcomes of the subjective evaluations. However, multiple iterations of the proposed algorithm increased the PSNR quality by up to 12 dB after the first three iterations. By comparison, the proposed algorithm significantly outperformed existing image enhancement techniques such as Histogram Equalization, Multiscale Retinex, and Low-light Image Enhancement via Illumination Map Estimation.