篇名 | Superpixel Segmentation Using Improved Lazy Random Walk Framework Based on Texture Complexities |
---|---|
卷期 | 30:5 |
作者 | Yi-Xuan Zhan 、 Chin-Han Shen 、 Hsu-Feng Hsiao |
頁次 | 252-267 |
關鍵字 | image segmentation 、 lazy random walk 、 superpixel 、 EI 、 MEDLINE 、 Scopus |
出刊日期 | 201910 |
DOI | 10.3966/199115992019103005020 |
Superpixel segmentation has been a very important pre-processing step of many computer vision applications. By grouping the pixels with similar data properties, the computation complexity can be reduced since the scale of data processing has been transformed from pixel level to region level. In this paper, an improved superpixel segmentation approach using a lazy random walk algorithm is proposed. Two major improvements are applied to obtain the better visualization results: center relocation and splitting strategy modification. The improved performance is confirmed with the subjective and objective performance comparison. In particular, the sizes of the produced superpixels depend on texture complexities of different regions which can be more appreciated.