BRISQUEs

BRISQUE, which stands for Blind Referenceless Image Spatial Quality Evaluator, is a no-reference image quality assessment algorithm. It is designed to predict the perceptual quality of an image without requiring a pristine reference image. BRISQUE operates by extracting a set of statistically derived features from the image itself and then using a support vector regression (SVR) model to predict a quality score. The features are based on the natural scene statistics of the image's local contrast.

The development of BRISQUE was motivated by the need for efficient and effective image quality evaluation in scenarios where a reference image is unavailable or impractical to obtain. This is common in applications like image compression, transmission, and storage, where modifications to the original image can alter its quality. BRISQUE aims to mimic human perception of image quality, focusing on distortions that are readily noticeable to viewers.

The algorithm first computes a difference-based representation of the image, often by taking differences between adjacent pixels. It then extracts features from this representation that capture local variations and their statistical properties. These features are typically invariant to certain image transformations like brightness and contrast changes. Finally, the trained SVR model maps these extracted features to a quality score, usually on a scale from 0 to 100, where lower scores indicate better perceived quality. BRISQUE has been widely adopted due to its simplicity, computational efficiency, and reasonable performance in predicting subjective image quality.