Implementation of Deep Learning Based Compression Technique and Comparative Analysis With Conventional Methodologies

Recently, the movement in information science and advancement completely impacts the human ordinary everyday practice. The size of mechanized data is filling rapidly to achieve images of unrivalled grade. The image compression works with to send large size images with insignificant bytes and to restore the image with incredible quality on social event. The focuses are to make an extra little data size; better nature of data during entertainment and connect transmission of data over confined information move limit with security. In this paper, conventional compression techniques i.e. Luminous DCT, Biorthogonal DWT, Quadtree Fractal and Huffman compression, are implemented and analyzed based on compression efficiency parameters like compression ratio, PSNR, MSE and SSIM. Here a deep learning based compression technique using stack autoencoder model is also proposed and implemented to compress the image data and result are compared with conventional compression techniques based on compression efficiency parameters. Simulation results show that proposed deep learning based compression scheme provides high quality reconstructed images having satisfactory compression ratio.