Content-Based Image Compression Using Hybrid Discrete Wavelet Transform with Block Vector Quantization

Image compression is necessary for the conveyance of information in the form of images. Images that have been compressed are fewer in size and sent over networks more quickly. Many algorithms focus on compressing images without prior knowledge on the image content type. But certain applications require content-based compression where degree of compression is controlled based on the image content type and should be able recover completely without loss of information. The proposed work aims at compressing the images based on the contrast variations and hybridizing discrete wavelet transform (DWT) and block vector quantization (BVQ) techniques. Two level DWT is applied on the image, then each sub-band is divided into non-overlapping blocks and a decision is made for each block based on the block variance before going for quantization. The proposed work calculates variance at the local regions to make decision as lower and higher contrast blocks, this helps to control degree of compression as only redundant/repeated blocks are allowed for quantization by preserving the edge information. Considering the entire image at once for vector quantization (VQ) diminishes the images' quality of compression. The VQ compression method often makes use of codebooks which possess lack of optimization. The proposed work implements BVQ technique, where only minimal pixels in a block are considered for quantization at once. This technique greatly reduces the computation time and also increases compression ratio. At last, huffman encoding is applied to the quantized coefficients. Following that, the bits that constitute the compressed image are saved and later restored. The suggested approach compresses and reconstructs images with adequate quality, on number of standard images as implemented. The effectiveness of the suggested work is also assessed by testing with custom real time images. When compared to current approaches, the findings suggest that the proposed work outperforms.