Research On Fast Globally Adaptive Dehazing And Quantitative Assessment Model

Fog is one of the main reasons that the image taken outdoors by vision technology would be degraded seriously.Nowadays,with the wide application of computer vision technology and the increasing influence of fogging weather on the quality of image,it is urgent to do more research on image dehazing.However,because of the complexity and uncertainty of removing the haze from image,although there exist a lot of methods and quantitative assessment models to restore the hazed image,they are not good enough and their effects are limited.Therefore it is significant to propose a fast,effective and scientific dehazing method and quantitative assessment model on image dehazing.This paper aims at solving the problem of dehazing globally and proposing a non-reference quantitative assessment model to evaluate the restored image.Firstly,the atmosphere scattering model and performance characteristics of quality degradation of hazed images are discussed.Then,the advantages and disadvantages of current dehazing methods are summarized and some tests are made.On the basis of existing research results,this paper proposes a fast globally adaptive dehazing algorithm and a quantitative assessment model based on visual perception.Finally,in accordance with application of the project,a series of software and hardware are developed and demonstration has been carried out.The tests showed that the method and the model proposed in this paper have the characteristics of effectiveness and robustness.Specific research work is carried out from the following four aspects:(1)Traditional algorithms for dehazing such as histogram equalization and Retinex and that based on atmosphere scatting model such as dark channel prior and Tarel method are analyzed in detail from the point of operation theory,handling procedure,realization effects,etc.Specially,the algorithm based on dark channel prior is emphasized and its application is analyzed and discussed.Finally,various algorithms are used to test their effects on globally dehazing image,and a comparative analysis of the effects and data of dehazing is made from the point of objective and subjective quality assessment as well as the speed of each algorithm.It is proved that every algorithm can provide some improvement of different extents for dehazing the image while the algorithm based on the dark channel prior is better than others.(2)Based on the limitations of current algorithms,a new fast adaptive globally dehazing algorithm(FGAD)is proposed and this algorithm is proved to be more scientific and effective.Gaussian pyramid and Laplacian pyramid are used to replace the traditional raw transmission map method,and it can avoid Halo Effect.While using Laplacian pyramid,we reserved the mapping function to restore the Original image,and then we use guided filter to refine raw transmission map.Moreover,we use filters to process brightness image of Original images to find the global air light map.Compared with the traditional method,FGAD can efficiently control Halo Effect.In order to suppress the over enhancement effect,a correction function of the fine transmission map is constructed to remedy the defect of the algorithm in bright field.Finally,we made a detailed and comparative analysis of the vision effects of the algorithm subjectively and objectively tested the algorithm,getting much data concerning image dehazing results,process time,etc.The experimental results showed that FGAD is as effective as DCP in dealing with thin foggy image,and is better than DCP in thick foggy image.More efficient than DPC in dehazing images in small sizes,while faster than DCP in dehazing images in big sizes.(3)A non-reference quantitative assessment model(QAMVP)based on human visual system is proposed.To avoid the problem that the dehazing effects could not be assessed completely while using a single parameter,we explored human visual system,optimize the most sensitive parameter for our assessment model.Five parameters,edge-preserving,hue recovery,structure information,MS-SSIM and VIF,are integrated to build an assessment model based on visual perception.The advantages of QAMVP is that by imitating human visual perception,it can avoid the drawback of the traditional non-reference quantitative assessment model with a single parameter and transform subjective assessment to mathematical question.The experimental results showed that the assessment model proposed in this paper is more efficient than traditional assessments in evaluating dehazed images.(4)The hardware and software of present algorithm are carried out in C++ language and processed in Visual Studio.The software that called test platform is used to test the algorithm and show the information between before and after dehazing.The main screen of test platform shows Original image and clear image at the same time so we can easily tell which algorithm is better than others.And it can also show details of the algorithms that help us do a variety of tests.To explore the application of our algorithm,using CCD,DSP,etc,we build up a test hardware based on TMS320DM642.After that we improved the test hardware by TMS320DM8168.The experimental results showed that it can process the hazy image and brings in an excellent result of realtime dehazing.Finally,the application demonstration of the industrial terminal was completed in the project demonstration area.