Research On The Polarimetric Image Dehazing Technique

In this dissertation,based on the need of a National Natural Science Foundation of China and a National High Technology Research and Development Program of China,research into the problem of image degradation under hazy weather is conducted,the influence of haze particles on the incident light is analyzed and the model of image degradation under atmospheric scattering theory is built.Based on the study of the retrieval technology of polarization information and the implementation method of imaging circuit,a polarization imaging system which has capability to acquire the complete Stokes parameters of the scene is designed.Under hazy weather,based on the polarization difference of the direct light reflected from the scene and the airlight scattered from the haze particles,an improved polarimetric dehazing algorithm is proposed to reverse the process of image degradation.Based on the spectral difference of image degradation level under hazy weather,a method which combines the polarimetric dehazing method and the spectral analysis technology by expending the process objects from ordinary visible light images to spectral images is proposed,the result of image restoration is enhanced through the realization of multi spectral image dehazing.The main contents and innovative points are:1.An imaging system which has the capability of acquiring full Stokes parameters is designed.The imaging circuit of the camera is designed based on the multi-aperture polarization optic system to acquire full Stokes parameters of the scene.The original images processed by the algorithm is changed from “the best” and “the worst” images to fixed polarization images,all the raw data needed by the dehazing algorithm can be captured in one shoot after the improvement.The polarization camera has high flexibility,two data reconstruction algorithms are realized in single control logic which realizes double AD transition online;the maximum theoretical frame rate of the sensor is transcended by interleaved sampling which realizes high frame rate;target tracking algorithms can run inside the camera which realizes the output of the area of interest of the scene.2.An improved polarimetric image dehazing algorithm is proposed.The global parameters of the polarimetric dehazing algorithm are estimated from the sky area.An image extraction method of sky area base on the physical property of the scene is proposed.Not like the traditional image-procession-based methods,this method utilizes the by-product of the image dehazing algorithm to realize the sky extraction which reduces the complexity of the algorithm and increases the stability of the dehazing result.The multi-scattering effect is not considered in the atmospheric scattering model used by this algorithm.Based on the relationship between the image contrast and the dehazing result,a self-adaptive revising method is proposed to revise the global parameter of the algorithm.In a set of dehazing results,the dynamic range and the contrast of the image are increased by 36.3% and 38.5%.3.The image restoration result of the polarimetric dehazing algorithm is enhanced by spectral analysis technique.The spectral analysis technology is applied to the image dehazing method.Based on the dehazing experiments under different spectral channels and the data analysis of the dehazing results,the relationship between the image restoration result and the spectral channels is given which is the experimental foundation of the polarimetric spectral dehazing method.In order to realize fast self-adaptive processing of the multi-spectral images,an improved polarimetric image dehazing algorithm which uses the angle of polarization as the main parameter is proposed.This algorithm requires no sky area in the original images which can expend the application area of this dehazing method.When applying the traditional dehazing methods to color images,serious color distortion can be found in the dehazing results,the cause for that problem is analyzed.The RGB channels of the dehazing result is balanced through multi-spectral processing which fixes the color distortion.The amount of color distortion is introduced to make quantitative analysis of the color distortion,in a set of dehazing results,our method can reduce the amount of blue-shift by 59%.