针对水下图像对比度低、边缘模糊、噪声大等特点，提出一种基于非下采样Contourlet变换和多尺度Retinex的水下图像增强算法。将水下图像进行多尺度多方向的非下采样Contourlet变换;利用多尺度Retinex算法调整低频系数，提高图像整体对比度;在各带通方向子带上估计噪声，抑制模值小于阈值的系数，改进神经网络中的Sigmoid函数用于调节模值大于阈值的系数;经非下采样Contourlet逆变换得到增强图像。与几种传统增强算法相比，本算法处理的图像达到了抑制噪声、改善图像对比度、突出目标轮廓的目的，具有较高的对比度评估值。

Considering the low contrast,fuzzy edge and strong noise of underwater images,an underwater image enhancement algorithm based on nonsubsampled Contourlet transform (NSCT) and multi-scale retinex is proposed. The underwater image is decomposed by nonsubsampled Contourlet transform. Then multi-scale Retinex (MSR) is used to enhance the image’s global contrast in the low frequency subband,and an improved Sigmoid function is used to process the coefficients at each scale in the directional bandpass subbands respectively,which suppresses small coefficients and enhances large coefficients by threshold denoising method. Finally,the enhanced coefficients are reconstructed to obtain enhanced image. Compared with the other traditional methods,the approach reduces the noise of original images,improves the contrast of the image,enhances object’s contour and achieves higher quantitative contrast assessment indicator.