天空区域分割的暗原色先验去雾视觉优化技术

Visual Optimizing Technology of Dark Channel Prior Dehazing Based on Sky Region Segmentation

摘要:

暗原色先验规律在天空区域的不适用,将会导致去雾后图像中的天空区域产生明显的噪声放大和色彩失真,为此提出基于天空区域分割的改进暗原色先验去雾算法。首先,采用将K均值聚类与增强边缘提取相结合的方法来进行天空区域分割,之后对有雾图像中天空区域的透射率进行修正,以得到改进的去雾后图像。该方法在天空区域分割的准确性上较好,去雾后图像不仅天空区域失真与噪声等显著减弱,还保证了远景清晰度。实验表明,该方法明显改善了去雾后图像天空区域的视觉效果并保留了远景清晰度,使去雾后图像显得清晰的同时表现得更加自然。

Abstract:

The failure of dark channel prior in sky region will lead to obvious noise amplification and color distortion in sky region of defogging images. Thus an improved dark channel prior dehazing algorithm based on sky region segmentation is proposed. Firstly, a method combining K-means clustering with enhanced edge extraction is proposed for sky region segmentation. After that, the transmission values in the sky region of the hazy image will be corrected to obtain improved dehazed images. The proposed method has a higher accuracy on the result of the sky region segmentation. Besides, it can weaken the distortion and noise in the sky region. At the same time, this method can also guarantee the sharpness of the distant views. Experimental results show that the proposed method can significantly improve the visual effects in the sky region of dehazed images. And meanwhile, it can keep the sharpness of the distant views. In summary, the proposed method can make dehazed images more clear and natural.