JPEG Discrete Cosine Transform Explained | Image Compression

Added:

Discrete Wavelet Transform (DWT) and Multi-Resolution Compression

While the Discrete Cosine Transform (DCT) is the foundation of traditional JPEG compression, it suffers from noticeable "blocking artifacts" at high compression ratios because it processes images in isolated 8x8 pixel blocks. A major alternative is the Discrete Wavelet Transform (DWT), utilized in the JPEG 2000 standard. Unlike DCT, DWT analyzes the entire image globally at multiple scales (multi-resolution analysis). This eliminates blocking artifacts, resulting in smoother image degradation at low bitrates and supporting progressive transmission, where an image refines from blurry to sharp. Furthermore, modern advancements have introduced Neural/Learned Image Compression, which uses deep learning autoencoders to bypass traditional mathematical transforms entirely.

Basic understanding of digital images, including pixel representation and color spaces like RGB and YCbCr.
The fundamental differences between lossy and lossless data compression concepts.
Introductory linear algebra, specifically matrix operations, as the DCT operates on 8x8 blocks of pixel values.
The concept of spatial frequency and the basic principles of transforming signals from the spatial domain to the frequency domain.
The subsequent stages of the JPEG pipeline: Quantization (discarding high-frequency data) and Entropy Coding (Huffman and Run-Length Encoding).
Analyzing compression artifacts, specifically the 'blocking artifacts' that occur at low bitrates due to the 8x8 DCT grid.
An exploration of the Discrete Wavelet Transform (DWT) and how it is utilized in the JPEG 2000 standard to overcome the limitations of DCT.
Application of DCT in modern video compression standards such as H.264 (AVC) and H.265 (HEVC).
The Modified Discrete Cosine Transform (MDCT) and its role in perceptual audio compression standards like MP3 and AAC.
679.1K views15.9Klikes15:11@ComputerphileOriginal Release: 2015-05-22

JPEG compression works by converting an RGB image to YCbCr color space, then dividing it into 8x8 pixel blocks, applying the Discrete Cosine Transform (DCT) to represent each block as a sum of cosine waves with different frequencies, calculating coefficients for each wave, quantizing these coefficients to discard high-frequency information (which humans perceive less), and finally using Huffman encoding to compress the remaining data efficiently.