[Wallace, 2001. pp3-6]

The JPEG compression algorithm involves three stages – Colour Space Conversion (Red-Green-Blue triplets mapped to Luminance-Chrominance Blue-Chrominance Red triplets), Segmentation into 8×8 pixel blocks, and Discrete Cosine Transformation (image transformation from spatial domain to frequency domain representation) encoding stage that includes Quantization (sorting the output waveform of DCT), Zigzag Scan (sorting the frequency coefficients from low to high), and Entropy Coding (Huffman Coding and Arithmetic Coding).

The encoding and decoding algorithms for JPEG images are shown in figures 1 and 2 respectively. Compression is used to reduce the file sizes such that they can be used on web pages or documentation or any other light weight graphics applications. The 8×8 blocks of source images are first shifted from unsigned integers to signed integers and then are applied to the Forward Discrete Cosine Transformer (FDCT). To decode the compressed image to achieve the 8×8 blocks again, they are applied to the entropy decoder whereby the output is taken out from Inverse Discrete Cosine Transformer (IDCT). The final processing step of DCT encoder is the entropy coding that achieves additional compression by encoding the quantized DCT coefficients based on statistical characteristics. There are two types of Entropy Coding methods – Huffman coding and Arithmetic coding. Huffman coding requires one or more sets of Huffman code tables whereas Arithmetic coding doesn’t require any external tables (although statistical conditioning tables as inputs can improve coding efficiency). The quantization stage of JPEG algorithm results in loss of information, thus making JPEG standard a lossy compression technique. [Wallace, 2001. pp3-6]

Compression Ratios and JPEG image formats:

Compression is specified in terms of bits per pixel including the chrominance and luminance components) called the compression bit rate. Higher the value of bits per pixel, better would be the image quality. The JPEG standard recommends the following mapping of image quality with bit rates in colored images having moderately complex scenes:

0.25 to 0.5 bits per pixel: The output image shall possess moderate to good quality

0.5 to 0.75 bits per pixel: The output image shall possess good to very good quality

0.75 to 1.5 bits per pixel: The output image shall possess excellent quality

1.5 to 2.0 bits per pixel: The output image shall be practically non-distinguishable from the original image

The improved version of original JPEG technology is JPEG 2000 that provides better rate

Computer Graphics