Design of a Haze Detection Algorithm and Hardware Implementation of an Image Dehazing Method

• Main Authors: Bo-Yi Wu, 吳柏毅
• Other Authors: Yeu-Horng Shiau
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/5y558p

碩士 === 國立雲林科技大學 === 電機工程系 === 103 === This study aimed to explore how to remove haze from a single image effectively. On the hazy weather, we can't obtain the clearly and completely image of the original scene. Thus, how to effectively remove haze from images until the objects covered with haze can be recognized by human vision would be the issue for needed discussion in image processing. In the future, if the haze removal algorithm is practically applied to surveillance products as monitors, vehicle video recorders, cameras for other uses, and so on, the amounts of calculation will be taken into account besides the good image quality. Therefore, we designed a haze detection algorithm that could clearly identify if haze was in images. If the haze detection algorithm and the haze removal algorithm were simultaneously applied to the surveillance systems, the haze detection algorithm would be used to identify if haze was in images in advance. If so, the haze removal algorithm would be carried out, and then the images would be outputted after processing. If not, we could output the images directly without the haze removal algorithm. Hence, we could leave out the processing time of the haze removal algorithm for non-haze images to achieve the purpose of lowering the amounts of calculation. This study proposed the haze removal algorithm mainly divided into two steps: first, use the dark channel prior and the mean filter to measure atmospheric light; second, use dark channel prior and the patch center point to determine if an edge existed and obtain the transmission values. With the atmospheric light and transmission values, we could employ the formula to restore haze-free image. The haze detection algorithm was divided into two steps: first, use the Hue of HSV color model to analyze the images and then segment the sky in images; second, use dark channel prior to analyze histogram intensity for identifying whether there was haze in images. Based on the proposed haze removal algorithm, we also designed a comprehensive hardware structure. Furthermore, the Verilog development tool we used is Altera’s Quartus II software version 11.0 SP1. In view of hardware design, the dark channel prior techniques of the above two steps would be simultaneously scanned. That would be beneficial to decrease the time and increase the efficiency of employing the haze removal algorithm.