Abstract: To overcome the problems in muti-exposure workpiece images, such as Low Dynamic Range (LDR), more reflective surfaces and excessive saturation, we propose a method of Retinex for enhancing the LDR image order, which extends the processing of Retinex on single image to image sequence and synthesizes the information from each image in the sequence. Introducing the illumination model, the proposed method solves the drawback that the traditional algorithm did not consider the influences of the illumination condition and the exposure time sufficiently. In addition, for protecting the edge, we decompose the approximate and detail components by the Nonsubsampled Contourlet Transform (NSCT), which has the band-pass character with multi-direction. Meanwhile, we separate the illumination and reflection components from the low approximate sub-image and select the optimal component combination. The results show that our proposed method not only removes the super-saturation of some reflective panels, but also promotes the visual dynamic range availably.

Key words: information processing technology, image enhancement, multi-exposure images, non-subsampled contourlet, Retinex