同态滤波耦合后处理优化的图像增强算法

目的为了提高图像增强算法中的亮度分布,抑制伪光晕与噪声现象,改善增强图像的细节与亮度。方法提出一种同态滤波耦合后处理优化的图像增强方案。首先,利用引导滤波器估计输入图像的照度,利用YCb Cr色彩空间中平滑的Y通道为导向图像,以有效捕捉真实场景的亮度,准确计算图像照度。其次,根据得到的照度估计,提取图像景物的反射率,为了同时提供动态范围压缩和色调再现,设计一种多尺度Retinex和色彩修复算子,利用3个不同尺度的Gaussian滤波加权组合,进行彩色图像增强。最后,为了实现多尺度Retinex与色彩修复算子的最优性能,通过非线性拉伸和参数优化组成的自动后处理方法,构建一种学习策略,利用量子行为粒子优化机制(QDPSO)自适应地确定每个输入图像的最佳参数,从而有效考虑了景物的照度与反射率的关系,避免了色彩失真。结果实验数据表明,与当前常用的增强算法对比,所提算法得到增强图像清晰度和细节更优,更符合视觉的感知特性,且效率更高,耗时约为0.7 s左右。结论所提算法具有良好的增强效果,在图像信息处理领域具有一定的借鉴作用。

The Image Enhancement Algorithm Based on Retinex Filter Coupling Post-processing Optimization

The work aims to improve the brightness distribution in the image enhancement algorithm, suppress the pseudo halo and noise, and improve the detail and brightness of enhancement image. An image enhancement scheme based on retinex filter coupling post-processing optimization was proposed. Firstly, a guided filter was used to estimate the illumination of the input image, and the smooth Y channel in the YCbCr color space was used as the guided image to effectively capture the brightness of the real scene and calculate the image illumination accurately. Secondly, according to the estimated illumination, the reflectivity of the image scene was extracted. In order to provide dynamic range compression and tone reproduction at the same time, a multi-scale Retinex and color repair operator was designed, which used three different scales of Gaussian filter weighting combination to enhance color image. Finally, to achieve the optimal performance of multi-scale Retinex and color repair operators, a learning strategy was explored through the automatic post-processing method consisting of nonlinear stretching and parameter optimization. The optimal parameters of each input image were determined adaptively by quantum particle swarm optimization (QDPSO), thus effectively considering the relationship between the illumination and the reflectivity of the scene, and avoiding the color distortion. The experimental data showed that, compared with the current commonly used enhancement algorithms, the proposed algorithm obtained higher clarity and details of the enhanced image, which was more aligned with the visual perception characteristics, and its efficiency was higher. The time consumed was about 0.7 s. With good enhancement effect, the proposed algorithm has certain reference value in the field of image information processing.