Optimized Deep Learning Approach for Image Compression Using Compressed Sensing Technique with Convolutional Autoencoders
Kiran Puttegowda, V. Veeraprathap, Sharmila Shanthi Sequeira, B Aruna, Kirankumarhumse, Sunil Kumar D S, K V Sudheesh · 2024
Image compression is crucial for efficient storage and transmission of visual data. Traditional methods like JPEG use transform coding, which may result in loss of fine details. Compressed sensing (CS) offers an alternative by exploiting signal sparsity, allowing for significant data reduction without substantial loss of information. Compressed sensing (CS) has proven to be an important technique in signal acquisition, especially in contexts in which sensor quality or the maximum possible duration of the measurement is limited. In this paper, deep learning techniques are used to improve compressive sensing in the context of image acquisition. In a existing approaches deployed stacked denoising autoencoders capable of reconstructing images considerably faster than conventional iterative methods. Apart from reviewing this approach, a possible extension using convolutional autoencoders inspired by the popular VGGnet architecture is discussed. Instead of learning models from scratch, a simple yet effective way for adapting available filters used in ImageNet classification is presented. By reformulation of the autoencoder structure in terms of a fully convolutional network can be adapted to arbitrarily large images for efficient learning of the measurement matrix and sparsity basis. Finally, experimental results indicate that the suggested approach outperforms current state-of-the-art methods across several metrics: reconstruction accuracy, time efficiency, and noise reduction capability. Particularly noteworthy is its exceptional reconstruction accuracy even with a limited number of measurements.