Prediction-based Lower Triangular Transform And Its Applications in Image Coding

碩士 === 國立臺灣大學 === 電信工程學研究所 === 88 === In order to replace the expensive DCT in JPEG standard, a new two-dimensional transform named HINT (Hierarchical INTerpolation) coder is proposed. This HINT coder is a special case of PLT (Prediction-based Lower triangular Transform) that is proved to have maxim...

Full description

Bibliographic Details
Main Authors: Chu-Ching Yang, 楊志清
Other Authors: See-May Phoong
Format: Others
Language:en_US
Published: 2000
Online Access:http://ndltd.ncl.edu.tw/handle/31574694087928057925
Description
Summary:碩士 === 國立臺灣大學 === 電信工程學研究所 === 88 === In order to replace the expensive DCT in JPEG standard, a new two-dimensional transform named HINT (Hierarchical INTerpolation) coder is proposed. This HINT coder is a special case of PLT (Prediction-based Lower triangular Transform) that is proved to have maximal coding gain as the unitary optimal KLT. The HINT coder is embedded in the JPEG by replacing the DCT and it will be shown to have nice performance. Since the four encoding modes in JPEG are carried out by four different programs, it’s not convenient and friendly for users. The proposed HINT-based JPEG can reach four modes at the same program and its progressive and hierarchical modes reduce lots of browsing time for Internet applications. In order to fit the different characteristics for the different parts of the same image, the predictors we used can be updated by the LMS method. A modification from the famous CALIC (Context-based, Adaptive, Lossless Image Coding) named GAP (Gradient-Adjusted Prediction) is used to make prediction according to the gradients of the current pixel. The GAP used in the differential coding system has comparable performance to the HINT coder for high bit rate coding, however, its performance in low bit rate is degraded. An interesting research is the incorporation of HVS (Human Vision System) into the differential coding system by some low pass and high pass filters.