Research And Implementation Of Non-Local Prior On Image Dehazing

For the past few years,the frequent occurrence of fog and haze has caused low visibility on roads,which fiercely affects traffic monitoring and meanwhile raises the traffic risk.Research on image dehazing of traffic video surveillance becomes a critical issue either on theory or application.This thesis focuses on the single image dehazing method on non-local prior,in reference of study on the dehazing method of physical model,and accordingly dehazing effects are promoted especially in foggy days.The specific research works are as follows:First of all,the general procedure of haze removal from image is studied for a better understanding of it's theory.Detailed analysis of atmospheric scattering model and image degradation model is useful for studying their principle and process one by one based on the classification of image enhancement or physical model with some classic methods.It proves that the non-local prior dehazing method based on the physical model makes algorithm efficiency higher and the dehazed picture more natural.Next,a further research on the dehazing method of single foggy image on non-local prior shows some problems with a particular analysis on the algorithm,like issue of initial transmissivity accuracy with cluster estimation by color value index and rough edge after image dehazing.Herein with the dehazing analysis on theory and experiments,transmissivity accuracy is improved by weighting context regularization of L1 norm to build the objective function for specifying the initial transmissivity.Furthermore,the image restoration is successful if the rough edge of dehazed image can be fixed with optimization of the specified transmissivity in domain transform filter for edge preserving.By using this method,it makes the dehazed image edge get as closer to the original and the color more vivid when experimentally verifying the dataset of general camera pictures and the traffic video image in foggy days,compared to the traditional non-local prior method.Meanwhile the image gets more natural from the subjective visual sense.Therefore,the objective evaluation index gets raised.Last but not least,a dehazing system of single image should be created and implemented to present the whole dehazing process and its effect.A visible interface can be made for a better presentation of the dehazing process on the GUI interfacefunction of Matlab.In addition,the quantitative analysis of the dehazed image can be acquired.