Resiliency of the Multiscale Retinex Image Enhancement Algorithm

The multiscale retinex with color restoration (MSRCR) continues to prove itself in extensive testing to be a very versatile automatic image enhancement algorithm that simultaneously provides dynamic range compression, color constancy, and color rendition. However, issues remain with regard to the resiliency of the MSRCR to different image sources and arbitrary image manipulations which may have been applied prior to retinex processing. In this paper we define these areas of concern, provide experimental results, and, examine the effects of commonly occurring image manipulations on retinex performance. In virtually all cases the MSRCR is highly resilient to the effects of both the image-source variations and commonly encountered prior image-processing. Significant artifacts are primarily observed for the case of selective color channel clipping in large dark zones in an image. These issues are of concern in the processing of digital image archives and other applications where there is neither control over the image acquisition process, nor knowledge about any processing done on the data beforehand. Introduction The Multiscale Retinex1 (MSR) is a generalization of the single-scale retinex2 (SSR), which, in turn, is based upon the last version of Land’s center/surround retinex3. The current version, the multiscale retinex with color restoration (MSRCR), combines the dynamic range compression and color constancy of the MSR with a color ‘restoration’ filter that provides excellent color rendition4 6. The MSRCR has been tested on a very large suite of images. However, concerns about its resiliency to both artifacts owing to digital image formation, and, to the digital processing performed on the image prior to the application of the MSRCR need to be addressed. We provide a general overview of the types of operations that can be performed on the image prior to dissemination and discuss their effect on the MSRCR output. Resiliency Webster’s Collegiate Dictionary defines resiliency as the “ability to to recover from or adjust easily to misfortune or change.” We have applied the MSRCR to images where we have no information either about the process that was used to form the image, or about any processing algorithms that were applied to the image. Resiliency in this context refers to the ability of the MSRCR to produce good (visual) images regardless of the characteristics of input image. Figure 1 shows the original image that we use throughout this paper and the MSRCR output using 4 scales. Though there appears to be a “graying-out” of the bright areas when compared with the original image, the sharpness and visibility of detail in the MSRCR output, more than compensate for any lack of local contrast. We use this original image and pre-process it to to simulate the commonly applied “enhancement” filters. Results are shown later in the paper. Multiscale Retinex with Color Restoration The general form of the MSRCR can be summarized by the following equation: ! "$# &%('*),+ -. / & 0 !132 (1) %4'5),+ 0687 & 0 !1 92;: =<> @?5 .A4A4A( B where is the < th band of the MSRCR output, C is the number of scales being used, # is the weight of the scale, is the < th band of the input image, and B is the number of bands in the input image. The surround function 7 is defined by 7 & 0 / D E GFIH JDKMLONQP