Line-based self-referencing string prediction technique for screen content coding in AVS3

String Prediction (SP) is a very efficient screen content coding (SCC) tool. In SP, the self-referencing string plays an important role to improve coding efficiency. But general self-referencing string has the problem of very low pixel copying throughput and is prohibited in the non-self-referencing based SP which has been adopted in the third-generation Audio Video Standard (AVS3). To overcome the problem and bring back the coding gain of self-referencing string, a line-based self-referencing string (LSRS) enabled SP technique is proposed. Moreover, to keep the pixel copying throughput and coding complexity of LSRS enabled SP the same as non-self-referencing based SP, an unbroken-line decomposition algorithm is presented to decompose an LSRS into multiple non-self-referencing strings. In this way, LSRS can be treated in the same way as a non-self-referencing string with the best trade-off between coding efficiency and complexity. Compared with non-self-referencing based SP, using AVS3 reference software HPM, for twelve SCC common test condition YUV test sequences in text and graphics with motion category and mixed content category, the proposed LSRS technique achieves the average Y BD-rate reduction of 0.81% and 0.59% as well as the maximum Y BD-rate reduction of 2.04% and 1.31% for All Intra and Low Delay configurations, respectively, with almost no additional encoding and decoding complexity. The proposed LSRS enabled SP technique has been adopted in AVS3 .