基于图像融合的水下图像清晰化方法

Underwater Image Clearness Method Based on Image Fusion

给出了推广的暗通道先验方法,由颜色的变化规律确定景深,利用景深信息提高背景光的估计精度。分别采用推广的暗通道先验、带有色彩恢复的多尺度Retinex以及对比度受限自适应直方图均衡3种方法对水下图像进行清晰化处理,将获得的3个图像作为图像融合的输入图像。综合考虑图像的可见度、饱和度、对比度和显著度等性能指标定义3个输入图像的权重图,以权重图作为融合权值对输入图像进行多尺度融合。实验结果表明:该方法能够显著提高水下图像的清晰度,并且能够有效校正色偏。

A generalization of dark channel prior(GDCP)method was given.In this method,the depth of field was determined by the law of color variation,and this depth information was used to improve the estimation accuracy of the ambient light.The underwater image was cleared by three methods:GDCP,multiscale retinex with color restoration(MSRCR)and contrast limited adaptive histogram equalization(CLAHE).Three images were obtained as input images for image fusion.The weight maps of these input images were defined by considering four image performance indicators:visibility,saturation,contrast,and saliency.Then,the multiscale fusion of the input images was achieved with the weight maps as fusion weights.The experimental results show that the proposed method can significantly improve the clarity of underwater images,and can effectively correct the color cast.