为了实现对亮度不均匀和含有复杂背景自然图像的快速准确分割, 本文提出一种结合Retinex校正和显著性分析的主动轮廓分割模型。通过引入Retinex校正, 获取图像中物体本身的反射分量, 可以抑制由非均匀光照带来的亮度不均影响; 另一方面, 经过Retinex校正后的图像更加客观地反映图像的本质, 确保了后续显著性信息提取的精确性, 使其更具实际指导意义。将显著性信息引入到主动轮廓模型之中, 有助于对含有复杂背景的图像进行有效分割。结合Retinex校正和显著性信息, 得到新的主动轮廓模型能量方程, 运用水平集方法指导曲线演化, 达到图像分割的目的。通过实验分析, 本文提出的主动轮廓模型可以实现快速、有效及鲁棒的图像分割。在MSRA数据库上平均处理速度为4.277秒/幅, F均值达到0.735。

To achieve fast and accurate segmentation of natural images with intensity inhomogeneity and complicated backgrounds, an active contour model combined with Retinex correction and saliency analysis for image segmentation was proposed. Retinex correction was applied to obtain the reflection component of objects in images; this could suppress the influence of intensity inhomogeneity caused by nonuniform illumination. Moreover, the Retinex-corrected image reflected the image essence more objectively, ensuring the accuracy of subsequent salient information extraction and making it more practical and instructive. The introduction of saliency information into the active contour model was helpful for the effective segmentation of images with complex backgrounds. By combining Retinex correction and saliency information, a new active contour model energy equation was obtained, and the level set method was used to guide the curve evolution to achieve image segmentation. Through experimental analysis, the proposed method was proved to be fast, effective, and robust. The average processing time on the MSRA database is 4.277 s per image, and the average F value is 0.735.