Global discrete cosine transform for image compression

Abstract

This paper presents a new lossy image compression method based on an orthogonal transformation (Global Discrete Cosine Transform, GDCT) using an optimal data truncation combined with an entropy coding (Run Length Encoding and Huffman encoding). The proposed method can be efficiently implemented without significant increase of system complexity. Simulation results are provided and compared to other results from the literature. The results reveal that our method achieves significant performance in term of distortion (PSNR) versus compression ratio.

Random projection and orthonormality for lossy image compression

“…Recent work in the field has focused on dimensionality reduction techniques to provide a high degree of compaction [3][4][5][6][7][8][9][10][11]. These are lossy algorithms that are quantitatively shown to discard redundant and irrelevant information, however, still providing a subjectively acceptable recovered image.…”

“…Dimensionality reduction techniques have the capability of reducing high-dimensional data into a lower-dimensional model such that the properties (i.e., distance between samples) are preserved. Methods such as Principle Component Analysis (PCA) [3][4][5], Singular Value Decomposition (SVD) [6][7][8], and the Discrete Cosine Transform (DCT) [9][10][11] have shown themselves applicable to the image compression problem.…”

“…The DCT has also been use in image compression for multimedia, medical imagery, and remote sensing applications [9][10][11]. A global DCT uses data truncation combined with run-length and Huffman coding [9].…”

“…A global DCT uses data truncation combined with run-length and Huffman coding [9]. After globally transforming the image, data truncation discards low-energy coefficients and then followed by entropy encoding with run-length and Huffman techniques.…”

“…Both of these are also known as dimensionality reduction schemes making them relevant for comparison to RP. Since RP is performed against the entire image, the DCT version used is similar to the global approach taken in [9]. The DCT is applied to the whole image instead of the traditional coding of 8 · 8 blocks.…”