Objective Object Segmentation Visual Quality Evaluation: Quality Measure and Pooling Method

Ran Shi,Jing Ma,King Ngi Ngan,Jian Xiong,Tong Qiao

DOI: https://doi.org/10.1145/3491229

2022-01-01

Abstract:Objective object segmentation visual quality evaluation is an emergent member of the visual quality assessment family. It aims to develop an objective measure instead of a subjective survey to evaluate the object segmentation quality in agreement with human visual perception. It is an important benchmark for assessing and comparing the performances of object segmentation methods in terms of visual quality. Despite its essential role, sufficient study compared with other visual quality evaluation studies is still lacking. In this article, we propose a novel full-reference objective measure that includes a two-level single object segmentation visual quality measure and a pooling method for multiple object segmentation overall visual quality. The single object segmentation visual quality measure combines a pixel-level sub-measure and a region-level sub-measure for evaluating the similarity of area, shape, and object completeness between the segmentation result and the ground truth in terms of human visual perception. For the proposed multiple object segmentation overall visual quality pooling method, the rank of each object's segmentation quality as a novel factor is integrated into the weighted harmonic mean to evaluate the overall quality. To evaluate the performance of our proposed measure, we tested it on an object segmentation subjective visual quality assessment database. The experimental results demonstrate that our proposed two-level measure and pooling method with good robustness perform better in matching subjective assessments compared with other state-of-the-art objective measures.