A lightweight handcrafted feature-based selective attention network for blind image quality assessment

• 主要作者: Feng, HaoLin
• 其他作者: Lin Weisi
• 格式: Final Year Project
• 語言: English
• 出版: Nanyang Technological University 2023
• 主題: Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• 在線閱讀: https://hdl.handle.net/10356/166046
• 機構: Nanyang Technological University

This paper introduces a novel approach to Blind Image Quality Assessment (BIQA) by employing handcrafted features combined with a selective feature attention mechanism, drawing inspiration from the human visual system (HVS). This method aligns more closely with human perception of image quality, as it integrates knowledge from the HVS. The handcrafted features utilized in this method include the Laplace operator, Scharr filter, and Discrete Cosine Transform (DCT), which were selected for their ability to capture essential low-level image properties critical for subjective image quality assessment tasks. The Laplace operator serves as an edge detection tool, the Scharr filter is a derivative-based filter for identifying edges in images, and the DCT helps analyze the frequency content of images. Compared to conventional BIQA algorithms, this method demonstrates improved accuracy while maintaining low memory complexity, making it suitable for real-time applications in image processing. This is particularly valuable in resource-constrained contexts where memory utilization is a major concern. The efficacy of this approach was validated through experiments on four synthetic distortion IQA datasets, with results indicating that the proposed method surpasses traditional BIQA algorithms in performance. In summary, the innovative BIQA method incorporates handcrafted features and a selective feature attention mechanism, drawing from the Human Visual System (HVS) to enhance image quality evaluation accuracy while preserving low memory complexity.