Lossless Multiband Image Compression System Based On Interband Context-Matching Prediction Scheme

• Main Authors: Yi-Chen Chang, 張宜琛
• Other Authors: 黃穎聰
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/42117125328367174305

碩士 === 國立中興大學 === 電機工程學系所 === 96 === In this thesis. Interband coding techniques are investigated for effective compression of multiband images such as color images and hyperspectral images. A multiband lossless compression system based on context matching prediction schemes is proposed. The observation is that the interband correlation in multiband images is usually greater than the intraband. Based on different spectral correlations among neigh- boring bands, we develop an adaptive prediction scheme that can dynamically select either inter or intra band prediction subject to spectral corrections. Specifically, we use context information to find out interband correlation between neighboring bands and then feedback the prediction error from previous band for bias cancellation. The proposed algorithm explores both interband and intraband statistical redundancies in coding. The simulation results show that the employment of interband prediction ac- hieves significant compression gains over its counterparts relying solely on intraband prediction. Our algorithm is particularly useful for data compression of image sources in band-sequential format (BSQ) or band-interleaved-by-pixel(BIP) format. Its performance evaluation on various sets of multiband image shows that it can outperform interband CALIC scheme by 0.2bpp on average. It is also achieves better compression results than the state-of-the-art algorithms such that JPEGLS and JPEG2000.