Research Of Image Enhancement Algorithms Based On Retinex Technologe And Implementation

With the progress of modern society, nightly image has been widely used in people’s life and work. But this kind of image normally is very dark, not suit for observation. Retinex algorithm could process it properly, but will cost a lot of capability and time. This thesis will analysis and improve the algorithm and finish a low computation loads prototype.This thesis researched image enhancement, color space, human visual system, frequency and spatial domain algorithm. Compare important algorithm in image processing, include homomorphic filtering, histogram Enhancement and Retinex. Analysis of the advantages and disadvantages of each method, and determine which algorithm is suit to improve.An ameliorative of MSR (Multi-Scale Retinex) is proposed based on the result of each Retinex computation loads. It could decrease MSR’s computing power with unobvious influence on the effect. The computation loads could be used again at next optimization. Detail optimizing idea showed as below.First of all, consider computer characteristic, the estimation scale could change from circle to square. Based on this idea, could use data mining etc. technique to decide an irregular estimation scale to get better effect for further research. Using this method could refer to farther distant point with same number of pixel. Second, according to human visual feature, use average weight in small area to instead of the weight calculated by surrounding function. It could decrease computation load by saving a lot of multiplication, and the effect is similar with MSR result. The last but not least is processing less point with Retinex algorithm. Only part of pixel will use MSR to calculate. the others will use related pixel value, which is already calculated by MSR, to estimate brightness value. The difference between the two method are very small. Computation loads will save more than half.At the end of this thesis, analysis the improved method result and MSR result with subjective and four objective methods to evaluate. Objective methods include brightness, contrast ratio, information entropy, handling time. The improve method retain MSR’s advantage and reduce computation loads a lot. For using software to process Retinex provide a good foundation.