Abstract:
Classic image processing problems include lossy image compression (LIC), which uses inexact approximations to compress images. By learning an encoder-quantizer-decoder network from a lot of data, deep convolutional neural networks (CNNs) have produced interesting results in LIC.
Existing CNN-based LIC methods can only train a network for one bits-per-pixel (bpp). CNNs are limited to LIC applications by the “one network per bpp” problem.
This image processing project proposes learning a single CNN for LIC at multiple bpp rates. A simple but effective Tucker Decomposition Network (TDNet) with a novel TDL decomposes a latent image representation into projection matrices and a core tensor.
We can easily adjust the latent image representation bpp rate within a CNN by changing the core tensor rank and quantization. An iterative non-uniform quantization scheme optimizes the quantizer, and a coarse-to-fine training strategy reconstructs decompressed images. Extensive experiments show TDNet’s PSNR and MS-SSIM compression performance is top-notch. Read also Communication Projects.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
