Heterogeneous Image Knowledge Driven Visual Perception

This letter introduces a novel framework for visual perception, specifically tailored to address challenges posed by heterogeneous images. In the face of diverse image modalities, traditional computer vision algorithms often struggle to perform effectively. To overcome this limitation, Our framework is driven by both representation and knowledge, combining visual representation obtained through extraction and learning with knowable representation tailored to specific heterogeneous image tasks. Through case studies on tasks like single image dehazing and face photo-sketch synthesis, we demonstrate the efficacy of our approach, showing significant performance improvements over traditional methods. This framework not only provides a solution to the challenges posed by heterogeneous images but also lays the foundation for further advancements in visual perception models. we present a novel framework for visual perception, designed to tackle the unique challenges posed by heterogeneous images. Traditional computer vision algorithms often struggle to effectively process diverse image modalities due to their inherent differences. To address this issue, we propose a visual perception framework that incorporates heterogeneous image knowledge, integrating domain-specific insights associated with different image types into classical visual perception models.