<p>Aiming at the problem of insufficient fine-grained feature extraction and fusion of images in current image quality assessment methods, a novel no-reference image quality assessment model with comprehensive features representation is proposed. The model mainly consists of four parts: image combination in different color space, multi-layer perceptual feature extraction, multi-layer feature combination, and quality regression prediction. Firstly, the image combination module in different color space combines images in different color spaces. Then, the multi-layer perceptual feature extraction module uses the ResNet-50 network to extract multi-layer perceptual features from images, from low-level semantics to high-level semantics. After that, the multi-layer perceptual features are combined using multi-combination strategy. Finally, the extracted features are input into the quality regression prediction module to obtain the image quality score. The model is tested on three internationally recognized image quality assessment databases, CSIQ, TID2013, and KADID-10K, and the experimental results demonstrated that the proposed model achieves a high degree of consistency with human subjective quality scores and higher accuracy in image quality prediction than existing methods.</p>