Abstract

Info

Title:: Multi-object optimization approach for artificial vision based on instance segmentation and saliency detection

Author(s):: WANG Jing1; 2; LIU Jianyun1; HAN Yanling1; ZHOU Ruyan1; SHEN Xiaojing1; 1. College of Information Technology, Shanghai Ocean University, Shanghai 201306, China 2. Key Laboratory of Fisheries Information, Ministry of Agriculture, Shanghai 201306, China

Keywords:: Keywords: retinal prosthesis multi-object recognition Swin-Transformer hierarchical optimization expression

Abstract:: Abstract: All foreground objects are extracted and segmented through instance segmentation using Swin-Transformer, and the image saliency features of luminance, size and location are combined to construct a multi-feature fusion attention hierarchical computational model for simulating human visual attention mechanism. The suitable phosphene resolution and luminance are adopted for foreground objects of different importance levels to realize the hierarchical optimization using different stimulus coding strategies. The simulation experiments of artificial prosthesis vision show that the experimental subjects will achieve higher recognition accuracy and take less time to complete multi-object recognition when they adopt the proposed multi-target hierarchical optimization approach. Instance segmentation technique is used to cascade the phosphene coding for mimicking human visual selective attention, thereby enhancing the multi-object perception in complex scenes for visual prosthesis. The study provides a reference for the development and application of image information coding and optimization strategies for visual prosthesis.