Research On The Key Technologies Of Retinex Image Enhancement Method

Image enhancement refers to that the use of some particular methods in image tohighlight some useful information which is used for special requirements and practicalapplication. The Retinex Theory is the main line of this paper. This thesis studies the basicframe of Retinex Algorithm and has the deep and systematic research of some keytechnologies on Retinex Algorithm. The main content is as follows:1、This thesis analyzes the Retinex Theory deeply and comprehensively， then analyzesthe computational model of Retinex from the mathmatic point of view, and classifies theRetinex Algorithm in different strategies of the path selection. It elaborates the Randompath Retinex Algorithm, Iterative Retinex Algorithm and Central\Surround RetinexAlgorithm, then points out the lacks of these algorithms.2、According to Receptive Field Mechanism, this thesis proposes a new Retinexalgorithm based on human visual properties. We sample the pixels in the concentriccircles area which take pending point as the centre with the gaussian distribution,calculate the highlight of the samples, then calculate the difference of lightness withthe pending point to restore reflective component of the objects. The algorithm ofthis thesis solves the problem that Retinex algorithm based on path strongly dependon the path and can’t get as much information from the neighborhood pixels.Moreover, the problem that Centre/Surround Retinex have got bad global propertiescan also be escaped by this algorithm. The experimental results show that thealgorithm of this thesis has wonderful enhancement effect and color fidelity in theglobal and local properties. Moreover, it escapes the “halo artifacts”.3、Current Retinex algorithm applied in the foggy image enhancement use fixedfilter, which can’t adapt to the situation of various depth of field and atomization,thus this thesis presents a self-adaptive filter Retinex algorithm based on DarkChannel Prior model. The model of Dark Channel Prior reflects the information offield depth and distribution of atmosphere in fog image, inspired, we design anself-adaptive filter according to the local value of Dark Channel, which usedifferent filters in different depth of field and foggy area to estimate theillumination component of the image, to achieve the clarity of foggy image. Finally,compare the result of this thesis algorithm with the result of HE algorithm and theresult of fixed filter MSR algorithm using the subjective observation and objectivedata analysis method. The comparison shows that the result of new Retinex algorithm based on Dark Channel Prior has better detail of image and global effectthan the HE algorithm and fixed filter MSR algorithm.