基于OpenCV的近景图像增强和分割算法研究

Research on Enhancement and Segmentation Algorithms of Close-Range Image Based on OpenCV

本文通过对近景图像的对比度和亮度进行分析实现了自动选择图像增强算法的方法,分别对金字塔均值漂移分割算法、分水岭分割算法和GrabCut分割算法进行了对比研究。经实验表明,图像对比度太小、亮度太高或太低对近景图像分割都会造成不利的影响,伽马变换增强算法适合调整亮度较高的图像,直方图均衡化增强算法适合调整亮度较低的图像。分水岭分割算法和GrabCut分割算法不适合近景图像多类别的自动分割,而金字塔均值漂移分割算法是基于核密度梯度估计的无参数快速统计迭代算法,可实现类别自动分割,分割精度较高,而且利用建立金字塔结构可进行多尺度分割。本文的研究成果对自动选择最佳的图像增强算法和分割算法以及参数设置具有重要的参考价值。

The method of automatically selecting the image enhancement algorithms based on analysis of contrast and brightness of close-range images was researched in this paper. A comparative study of pyramid mean shift segmentation algorithm, watershed segmentation algorithm and GrabCut segmentation algorithm was conducted. The results show that too small contrast, too high or too low brightness of images will adversely affect the segmentation effect of close-range images. The gamma transformation enhancement algorithm is suitable for adjusting the images with higher brightness, and the histogram equalization enhancement algorithm is suitable for adjusting the images with lower brightness. The watershed segmentation algorithm and GrabCut segmentation algorithm are not suitable for automatic multi-class segmentation for close-range images. The pyramid mean-shift segmentation algorithm is a non-parameter statistical iterative algorithm based on kernel density gradient estimation, which can be used to automatically segment images into multi-class patches, and segmentation accuracy is very high. Furthermore, multi-scale segmentation can be performed with this method. The research results in this paper have important reference values for automatically selecting the best image enhancement algorithm and segmentation algorithm, and setting parameters.