Image Compression Using Discrete Wavelet Transform and Convolution Neural Networks

• Published in: Journal of electrical engineering & technology 2024, 19(6), , pp.3713-3721
• Main Authors: Kumar, Gottapu Santosh, Rani, M. Laxmi Prasanna
• Format: Article
• Language: English
• Subjects: Electrical Engineering; Electrical Machines and Networks; Electronics and Microelectronics; Engineering; Instrumentation; Original Article; Power Electronics; 전기공학
• ISSN: 1975-0102; 2093-7423
• DOI: 10.1007/s42835-024-01803-0

The amount of information is growing very fast and multimedia files are the elements of information that occupy more storage. This problem calls for more efficient image compression techniques to save the cost on storage and transmission. Compression methods are used to convert the image files with less memory space compared to the original image. Transform based image compression has its significance in image compression but combining it with the developing technologies like deep learning produce efficient results. A lossy compression technique is proposed in this paper which incorporates Convolutional Neural Networks (CNNs) to predict wavelet high frequency coefficients from low frequency coefficients. The main premise of the proposed framework is that information which can be recovered at the decoder via CNN prediction can be excluded from the encoding bit stream that resulting in reduced size. This method of image compression is efficient compared with the individual methods of transform based and CNN based image compression. From the results, it is observed high PSNR of 42.688 dB, low MSE of 3.5015, high entropy and optimum SSIM of 0.986 is obtained using proposed model by considering X-ray image of chest with compression ratio of 66.351%.