Blind Image Quality Assessment in JPEG Compressed Domain

The perceptual quality of image and video data is of prime importance in mainstream and social media applications. Subjective quality assessment of image data is quite a task as the number of image data being shared on social media and other such platforms is exponentially increasing. Therefore, developing blind image quality assessment (BIQA) algorithms is crucial. A JPEG compressed domain BIQA method is proposed in this paper that works solely on the images in the compressed form rather than decompressing them like the traditional BIQA methods. The quantized discrete cosine transform (DCT) coefficients are obtained by entropy decoding and inverse quantization of the JPEG bitstream. Feature-maps-based Reference less Image Quality Evaluation Engine (FRIQUEE) based features are extracted from quantized DCT coefficients, which are then passed to support vector regressor for the quality score prediction. Results on both synthetic and authentic distorted datasets prove the effectiveness of the proposed technique. The computation cost of the proposed BIQA technique is less than conventional methods and a top SROCC score of 0.96 on benchmark datasets has been achieved.