Efficient Encoding of Wireless Capsule Endoscopy Images Using Direct Compression of Colour Filter Array Images

Abstract

Since its invention in 2001, wireless capsule endoscopy (WCE) has played an important role in the endoscopic examination of the gastrointestinal tract. During this period, WCE has undergone tremendous advances in technology, making it the first-line modality for diseases from bleeding to cancer in the small-bowel. Current research efforts are focused on evolving WCE to include functionality such as drug delivery, biopsy, and active locomotion. For the integration of these functionalities into WCE, two critical prerequisites are the image quality enhancement and the power consumption reduction. An efficient image compression solution is required to retain the highest image quality while reducing the transmission power. The issue is more challenging due to the fact that image sensors in WCE capture images in Bayer Colour filter array (CFA) format. Therefore, standard compression engines provide inferior compression performance. The focus of this thesis is to design an optimized image compression pipeline to encode the capsule endoscopic (CE) image efficiently in CFA format. To this end, this thesis proposes two image compression schemes. First, a lossless image compression algorithm is proposed consisting of an optimum reversible colour transformation, a low complexity prediction model, a corner clipping mechanism and a single context adaptive Golomb-Rice entropy encoder. The derivation of colour transformation that provides the best performance for a given prediction model is considered as an optimization problem. The low complexity prediction model works in raster order fashion and requires no buffer memory. The application of colour transformation yields lower inter-colour correlation and allows the efficient independent encoding of the colour components. The second compression scheme in this thesis is a lossy compression algorithm with a integer discrete cosine transformation at its core. Using the statistics obtained from a large dataset of CE image, an optimum colour transformation is derived using the principal component analysis (PCA). The transformed coefficients are quantized using optimized quantization table, which was designed with a focus to discard medically irrelevant information. A fast demosaicking algorithm is developed to reconstruct the colour image from the lossy CFA image in the decoder. Extensive experiments and comparisons with state-of-the-art lossless image compression methods establish the superiority of the proposed compression methods as simple and efficient image compression algorithm. The lossless algorithm can transmit the image in a lossless manner within the available bandwidth. On the other hand, performance evaluation of lossy compression algorithm indicates that it can deliver high quality images at low transmission power and low computation costs.