Quality prediction of compressed images via classification

Abstract

In this paper, we have investigated an image classification problem according to image quality after compression. A classificationbased image compression approach has been proposed, where images are assigned to one of two classes before their compression by a JPEG algorithm. This classification allows to set the proper value of Quality Factor (QF) for each image, in such a way, to save storage space while maintaining sufficiently high image quality. The image quality has been evaluated by a Structural Similarity (SSIM) index metric and Peak Signal-to-Noise Ratio (PSNR). As image classification results depend on the selected features describing the images, the feature selection problem has to be solved before classification. The experimental investigation has shown that the proposed approach allows to save storage space compared to a conventional JPEG algorithm. It is especially useful when saving huge amount of images.