Özet:

Abstract A large number of books/documents written in so many languages are available in digital libraries. These scanned copies of books/documents consume more space on hard disk.  If the existing image compression technique like JPEG is used for compressing the textual images then significant amount of information could be lost. Hence we proposed and implemented the Convolutional Autoencoder (CAE) for compression and decompression of handwritten Devanagari document images. As long as compression of document images is concerned, relatively large work was reported for documents having printed text. Comparatively less significant work was done on handwritten Devanagari document images. Convolutional autoencoders have been shown to outperform traditional compression methods like JPEG in terms of compression efficiency and image quality, but requires more computational resources. Convolutional autoencoders have the potential to become a widely adopted approach for image compression in the future, as computing power increases and the demand for higher quality compressed images grows. A lot of experimentation was done by the changing the compression ratios. The PSNR and MSSIM parameters are considered to estimate the performance of compression and decompression results. The compression and decompression results achieved with the help of CAE are much encouraging as compared to the conventional image compression.

Anahtar Kelimeler: