基于感知增强理论的视觉显著性计算

受人类视觉系统选择性注意机制启发的视觉计算方法是当前计算机视觉领域的研究热点之一。传统的视觉显著性计算模型大多是基于空间位置的，没有充分考虑图像的感知组织和结构信息，因而无法有效地应对许多实际应用中复杂视觉场景带来的挑战。本课题拟以视觉注意的感知增强理论为基础，研究和建立一种新的视觉显著性计算框架，即基于对象的视觉显著性计算。首先，按照感知增强理论所提出的“注意信息沿着格式塔线索传播”的基本结构，建立基于对象的视觉显著性计算的一般框架；其次，围绕着图像格式塔线索的提取与表达、注意信息传播模型这两个关键问题开展算法研究，提出具体的算法实现方案；最后，通过实验，并结合具体应用，对模型和算法进行性能评估和进一步优化完善。本课题从“基于空间位置”到“基于对象”，发展了图像视觉显著性计算的理论和方法，同时能够为视觉显著性计算在复杂视觉场景下的广泛应用提供算法支持。

Visual computing inspired by the selective attention mechanism of human visual system is at the frontier of computer vision. The vast majority of traditional visual saliency models are space-based, which do not take into account the perceptual organization and structural information of images, and are therefore unable to effectively cope with challenges posed by complex visual scenes in many real-world applications. This project aims to establish a novel computational framework for visual saliency, namely, object-based visual saliency, based on the Sensory Enhancement Theory of visual attention. Firstly, we will establish the generic computational framework of object-based visual saliency, which follows the paradigm of attention spreading along Gestalt cues as suggested by the Sensory Enhancement Theory. Secondly, we will concentrate on the two key fundamental issues: extraction and representation of Gestalt cues, and attention spreading, and propose the overall algorithmic solution. Finally, we will evaluate the performance of the proposed models and algorithms by experiments and applications, and make further improvements accordingly. Going from space-based visual saliency to object-based visual saliency, this project will promote the development of visual saliency computation, both theoretically and methodologically. It will also contribute by providing practical algorithms to enable the wide application of visual saliency in complex visual scenes.