Research And Implementation With Hardwares Of Dehazing Image Based On Retinex Algorithm

The frequently occurring of bad weather,such as fog and haze,has led to images degraded and visual quality declined in the field of aerial photography,remote sensing and traffic surveillance.In consideration of this problem,imaging enhancement is often used to recover and strengthen image in environment of haze or smoke.The application of many image enhancement methods is limited because these methods make images dehazing by using computer.How to improve the practical value of dehazing image algorithm has become the focus and difficulty in the field of image processing.Surrounding this subject,a FPGA+ARM image processing framework using multi-scale Retinex was designed and a hardware processing system with high real time and good treatment result was researched in this paper.The main research content includes: the selection and the improvement of image dehazing lagorithm,the hardware system design,the hardware implementation of the algorithm and the actual testing and data analysis of the hardware system.Firstly,the algorithms of dehazing image are introduced.The theoretical knowledge and implementation of multi-scale Retinex is discussed emphatically.Then the algorithm results of various image dehazing are campared and analysed.Contrast limited adaptive histgram equalization is used to replace the original algorithm of image stretching.In consideration of the results and the calculation,improved multi-scale Retinex is selected for hardware transplanted.Secondly,a FPGA+ARM image processing hardware system is designed on the basis of actual requirements of the improved multi-scale Retinex hardware transplanted.This hardware system has a good system management capabilities and excellent data parallel processing capabilities in which ARM is the control center and FPGA is the data processing center.In the design of circuit,stacked structure is conducive to change the image input interface board optionally.It not only can reduce volume of the system,but also expand application of the system.Finally,in the process of algorithm transplanted,the course of convolutional filtering is implemented and the process of large scale convolution filtering is simplified benefit from the FPGA's advantages of pipeline operation and parallel processing.These studies will benefit to reduce the consumption of hardware and improve the system's real-time ability.Besides,taking advantage of ARM's strengths that can run operating system,image enhancement is implemented via transplanting the algorithm of contrast limited adaptive histogram equalization.Experiments show that the system accomplish the task of multi-scale Retinex transplanted and has good real-time performance for 1024×1024,40 fps video dehazing with lower power consumption and smaller volume.The research and verification of improved multi-scale Retinex transplanted solve the problem in hardware implementation of object identification and image enhancement.It also play an important role in the field of aerial reconnaissance and video surveillance,at the same time,it provides a reference for the hardware thansplantation of other image enhancing algorthms.