Image Compression using ITT and ICT and a Novel Variable Quantization Technique

The need for novel transform coding techniques promising improved reconstruction and reduced computational complexity in the field of image and data compression is undeniable. Currently, there is a prevalent use of the integer adaptation of the popular discrete cosine transform (DCT) with fixed quantization in the video compression domain due to its ease of computation and adequate performance. However, there cannot be a single method or technique that would not only provide maximum compression possible, but also offer the best quality for different types of images. The influence of specific features of the image, such as its structure and content, on the quality of reconstructed image after decompression cannot be ignored. This thesis intends to utilize this aspect and identify areas where an alternative to integer DCT (ICT) for image compression can be proposed. There exist polynomial-based orthogonal transforms like discrete Tchebichef transform (DTT), which possess valuable properties like energy compaction, but are potentially unexploited in comparison with DCT. This thesis examines, in detail, where DTT stands as a viable alternative to DCT for lossy image compression based on various image quality parameters. It introduces a multiplier-free fast implementation of integer DTT (ITT) of size 8×8 that significantly reduces the computational complexity. Normally, images have detail spread across them in a non-homogenous manner. Hence, when the image is divided into blocks, some blocks might have intricate detail, whereas the amount of detail in some might be very sparse. This feature is exploited in this thesis by proposing a technique to adapt the quantization performed during compression according to the characteristics of the image block. The novelty of this variable quantization is that it is simple to implement without much computational or transmission overhead. The image compression performance of ITT and ICT, using both variable and fixed quantization, are evaluated and compared for a variety of images. Eventually, the cases suitable for ITT-based image compression employing variable quantization are identified