Near-lossless Image Compression with Parity Reduction

In our previous work, we introduced a lossless image compression algorithm (called BWIC_I) based on hierarchical prediction, inversion, and context-adaptive coding techniques. We showed that BWIC_I outperformed most commonly used standard color image compression algorithms, including JPEG in lossless mode and JPEG 2000, on a variety of well-known image data sets. In this study, we introduce a near-lossless image compression algorithm that essentially consists of parity reduction by dropping the least significant bit of pixel values of RGB color images and then compressing the resulting data with BWIC_I. The proposed algorithm provides a guaranteed minimum PSNR value. Moreover, it does not require a specialized encoding and decoding algorithm and can be utilized in conjunction with other generic image compression algorithms. The compression performance of the proposed technique is considerably better than JPEG and JPEG 2000, as demonstrated on the standard Kodak image set.