Canny SLIC to Compute Content-Sensitive Superpixels
Yousef Abu Baker, Iker Gondra · 2018
Superpixel segmentation is becoming a ubiquitous initial preprocessing step in computer vision applications. It is important that the resulting superpixels preserve image boundaries. The simple linear iterative clustering (SLIC) algorithm is a popular method for superpixel segmentation. However, in the case of content-sensitive superpixels, which are located in small structured -dense regions with high color variation, SLIC may not generate boundary-preserving superpixels. More complex methods have alleviated this problem by, e.g., using distance measures other than Euclidean distance. However, as an initial preprocessing step, the simplicity of superpixel segmentation is crucial. We propose a relatively simple method that uses the Canny edge detector in combination with SLIC to generate superpixels that tend to preserve image boundaries.