Improved AC Coefficients Compression Using Alternate Adaptive Scanning and Multi-Feature Determined Context Arithmetic Coding

For nearly three decades, JPEG has been the dominant method for compressing still images. Despite its simplicity, its significance remains undiminished. In this study, we compressed JPEG AC coefficients with different methods. The conventional zig-zag scan order was substituted with an alternative adaptive scanning order and determined by the likelihood of its neighboring coefficients being non-zero and the gradients calculated from the DC coefficients. While run/length was coded in JPEG, we encoded them separately to extract more information to reduce redundancies. Then, three types of contexts were determined by the position of diagonal entries, the mean value of the neighboring corresponding entries, and the run value. Additionally, we incorporated frequency table initialization, range adjustments, and adaptive bounding according to the concept of Context-Based Adaptive Variable-Length Code (CAVLC). The proposed method outperformed the JPEG by an 8.92% reduction in bpp on average.