| Summary: | 碩士 === 國立成功大學 === 電腦與通信工程研究所 === 101 === Images of outdoor scenes are usually degraded because the ambient light is scattered, especially in haze. Many details will be lost in hazy images and these images are inappropriate to be used in many applications such as surveillance, intelligent vehicles and object recognition. Fortunately, single image haze removal methods have been proposed recently. In this thesis, an improved single image dehazing algorithm is proposed which is based on dark channel. First, the proposed algorithm determines the color vector of the atmospheric light after averaging the brightest pixels. And it analyzes the brightness distribution of the hazy image to estimate the intensity of the atmospheric light simultaneously. As a result, it can avoid color cast and dim phenomenon in output images. Furthermore, it also removes the redundant high value pixels of the coarse transmission map to reduce the halo artifact. Finally, the proposed algorithm introduces an exponential weight when recovering the scene radiance, so the details of image can be clearer.
The experimental results show that it achieves as good or even better results compared to the main present-day algorithms as illustrated.
|
|---|
