Image enhancement using similarity-based color matching for high-quality stereoscopic 3D image acquisition

Abstract

Stereoscopic three-dimensional (S3D) movies often contain inconsistencies between left and right images acquired by a stereo camera due to an unstable filming environment. This research introduces a novel image enhancement algorithm using similarity-based color matching between S3D images. The proposed algorithm first partitions both reference and target images into multiple sub-blocks, and then classifies those blocks into reflection and illumination components using retinex theory. Color correction is performed by matching histograms of a corresponding pair of blocks based on the structural similarity index measure (SSIM). The color corrected images are further enhanced by removing noise in the wavelet transform domain. We can make high-quality S3D images from imperfect input images acquired under stressful conditions including limited dynamic range, unstable calibration of stereo camera pairs, and low signal-to-noise ratio (SNR). The proposed method can be applied to high-quality panorama images, frame difference-based video tracking, and similarity-based image analysis as well as S3D films.