A combined method for multi-class image semantic segmentation

摘要

Multi-class image semantic segmentation (MCISS) is one of the most crucial steps toward many applications related with consumer electronics fields such as image editing and content-based image retrieval. Existing MCISS approaches often consider only the top-down process and suffer from poor label consistency among neighboring pixels. To overcome this limitation, this paper proposes a combined MCISS method to integrate a state-of-the-art topdown (TD) approach Semantic Texton Forests (STF) and a classical bottom-up (BU) approach JSEG to exploit their relative merits. Experimental results on two challenging datasets show that the proposed method can achieve higher accuracy in comparison with the original STF method, while it does not notably prolong the computational time. In addition, several insights into the evaluation metrics of MCISS are reported.