Lifting-Based Block Fractional Wavelet Filter Compression of Hyperspectral Images over Wireless Multimedia Sensor Network Platforms

Abstract

In the rapidly development of remote sensing technology, the compression of Hyperspectral Images is a pivotal yet formidable task. Hindered by inherent limitations in hyperspectral imaging, enhancing the accuracy and efficiency of compression algorithm remains a critical and much-debated issue. Algorithms using set partition wavelet transforms excel in hyperspectral image compression due to their embedded nature, coding efficiency, and low complexity. Specifically, the Fractional wavelet-based zero memory set partitioned embedded block algorithm achieves high coding efficiency with lower memory demands, though its method of repeatedly comparing coefficients to a threshold is time-intensive. To solve this, a new algorithm has been developed that optimizes both computational and memory complexity. It employs a block-based fractional wavelet filter (BFrWF), which delivers the same accuracy as conventional transforms but requires far less memory.

The Block-based Fractional Wavelet Filter is a low-memory technique for image transformation, but its high computational complexity makes it impractical for resource-constrained devices in IoT and Wireless Sensor Networks. Additionally, it produces blocking artifacts due to improper handling of block boundaries. This paper introduces a new lifting-based version of BFrWF that eliminates these artifacts by correctly overlapping image blocks. This new implementation with low complexity zero memory set partitioned embedded block (LC-ZM-SPECK) achieves higher coding efficiency, making it well-suited for resource constraint visual sensor nodes.