annatz / jpeg-baseline-codec Goto Github PK

This project was developed for the 'Multimedia Systems and Virtual Reality' course, part of the MEng in Electrical and Computer Engineering at the Aristotle University of Thessaloniki in 2019. The aim of the project is to develop a JPEG encoder/decoder following the ISO/IEC 10918-1:1994 standard, focusing on the baseline sequential DCT-based process.

The primary goal is to assemble a library of functions that encapsulate key components of the JPEG encoding and decoding template. Each function will include its inverse to facilitate reconstruction, integral to the decoding process.

Initial preprocessing converts images from RGB to YCbCr format, supporting different subsampling formats (4:4:4, 4:2:2, and 4:2:0). The function signature is as follows:

function [imageY, imageCb, imageCr] = convert2ycbcr(imageRGB, subsampling)

• imageY, imageCb, imageCr: Components of the YCbCr image.
• imageRGB: The original RGB image.
• subsampling: A 1x3 array defining the subsampling format (e.g., [4 2 0]).

The inverse function, responsible for converting YCbCr back to RGB and oversampling to match the original image's dimensions, is specified as:

function imageRGB = convert2rgb(imageY, imageCr, imageCb, subsampling)

DCT transformations are applied at the block level:

function dctBlock = blockDCT(block)

And its inverse:

function block = iBlockDCT(dctBlock)

Quantization of DCT blocks and the corresponding inverse function are defined as:

function qBlock = quantizeJPEG(dctBlock, qTable, qScale)
function dctBlock = dequantizeJPEG(qBlock, qTable, qScale)

Functions for run-length encoding (RLE) and its inverse are outlined for processing quantized DCT coefficients:

function runSymbols = runLength(qBlock, DCpred)
function qBlock = irunLength(runSymbols, DCpred)

Huffman coding and decoding functions are provided for compressing run-length encoded symbols:

function huffStream = huffEnc(runSymbols)
function runSymbols = huffDec(huffStream)

demo1.m demonstrates the process of converting an RGB image to YCbCr and back, along with DCT coefficient processing. Different subsampling formats are used for two provided images.

Integrating the developed functions into a comprehensive encoder/decoder function is the second major deliverable, aiming at both quantitative and qualitative analysis of the achieved compression.

The function signatures for encoding and decoding are:

function JPEGenc = JPEGencode(img, subsampling, qScale)
function imgRec = JPEGdecode(JPEGenc)

demo2.m will calculate and compare entropies in different domains for two provided images, using the same parameters as in Demo 1.

The final deliverable focuses on constructing and decoding a bitstream of the encoded image, ensuring compatibility with standard image processing software.

The functions for encoding to and decoding from a JPEG bitstream are defined as:

function JPEGencStream = JPEGencodeStream(img, subsampling, qScale)
function imgCmp = JPEGdecodeStream(JPEGencStream)