Skip to Main Content
This paper explores the advanced features of JPEG 2000 and proposes the novel curvilinear-edge coding enhancement in the context of larger aerial image coding. This enhancement applies to the low resourceful on-board Machine Vision environment for the important surveillance applications. The overall memory consumption of the on-board codec's processor is lowered maximally by 20 percent with the use of the Precinct Partition strategy. Resolution prioritized progression ordering mode enables the immediate on-ground preview with undersized image reconstruction. Moreover, when coding larger aerial image's Region of Interest (ROI), JPEG 2000 with its layered architecture outperforms SPIHT. Using the average layer thresholds, a sequence of different aerial images can be reconstructed at the same quality. Exploring JPEG 2000 part II, i.e. 3-D Hyperspectral image coding concepts, the evaluated new parameter PSNR(ROI) (db) per band for mean bit rate of JPEG 2000 has a 5.7 percent gain over that with SPIHT. Despite its many powerful features, JPEG 2000 still lacks the directional detail filtering capability, hence, the inefficiency in curvilinear-edge reconstruction due to its inherent wavelet transforms based codec. Therefore, the novel Advanced Directional-Wavelet Packet Transform (ADWPT) is proposed for JPEG 2000, and implemented in pre-coding experiments giving a more competent curvilinear-edge coding performance than that with the existing transforms.
Machine Learning and Cybernetics, 2007 International Conference on (Volume:3 )
Date of Conference: 19-22 Aug. 2007