Skip to content

HW-Lee/ImageCodec

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

18 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ImageCodec

This is the final project of Video Signal Processing in NTHU, 2015, Spring, and the document is just a set of supplementary materials. It simply demonstrates how to use these source files to encode/decode an image, and how to integrate/compile the project. Because it is complex, I can only show how to use as clearly as possible. If you have any problem, please send me an e-mail. ([email protected])

Requirements

  • Mac OSX [recommended] or UNIX (terminal and shell script used)

  • Xcode-build-tools [recommended] or g++ compiler

  • Matlab [recommended] or YUV viewer

    Note that Windows has not been tested yet, so it might take lots of time encoding an image. Please wait for it, thanks.
    If you are trying to run this project on the platform other than Mac OSX, please make sure your executable (.out, .exe) is in the directory /ImageCodec/

Project Structure

  • Root: /ImageCodec/

  • Code: /ImageCodec/code/, /ImageCodec/libs/

  • Data: /ImageCodec/data/

  • Generated: /ImageCodec/results/

  • Report: /ImageCodec/report/

    Note that if user does not arrange files followed the structure mentioned above, it might cause some unknown errors.

Quick Start

Directly running the sample code

  • run cd ~/{YOUR_DIR}/ImageCodec/

  • Configuration in the sample code

    1. check the file /ImageCodec/code/ImageCodec_Demo.cpp

      #define CONSTRAINT_BY_PSNR 0
      #define CONSTRAINT_BY_BITRATE 1
      #define SKIP_ENCODING -1
      
      using namespace std;
      int main(int argc, const char * argv[]) {
      
      	int constWay = {Config}; // 
      						<-- can be
      						CONSTRAINT_BY_PSNR    : Use PSNR as constraint
      						CONSTRAINT_BY_BITRATE : Use bitrate as constraint
      						SKIP_ENCODING         : Skip encoding
      
      	int opt = {Config}; // <-- can be
      									0: 1_1536x1024.yuv
      									1: 2_1024x768.yuv
      									2: 3_1000x728.yuv
      									3: 4_1000x1504.yuv

      and

      	if (constWay == CONSTRAINT_BY_PSNR) {
      		ss << PATHS[opt+4]; ss << "bitstream1.bin";
      		ImageCodec::encodeTo(ss.str())->withImage(img)->withMinPSNR({Config})->run();
      		ss.str(""); ss.clear(); ss << PATHS[opt+4]; ss << "bitstream2.bin";
      		ImageCodec::encodeTo(ss.str())->withImage(img)->withMinPSNR({Config})->run();
      		ss.str(""); ss.clear(); ss << PATHS[opt+4]; ss << "bitstream3.bin";
      		ImageCodec::encodeTo(ss.str())->withImage(img)->withMinPSNR({Config})->run();
      	} else if (constWay == CONSTRAINT_BY_BITRATE) {
      		ss << PATHS[opt+4]; ss << "bitstream1.bin";
      		ImageCodec::encodeTo(ss.str())->withImage(img)->withMaxBitrate({Config})->run();
      		ss.str(""); ss.clear(); ss << PATHS[opt+4]; ss << "bitstream2.bin";
      		ImageCodec::encodeTo(ss.str())->withImage(img)->withMaxBitrate({Config})->run();
      		ss.str(""); ss.clear(); ss << PATHS[opt+4]; ss << "bitstream3.bin";
      		ImageCodec::encodeTo(ss.str())->withImage(img)->withMaxBitrate({Config})->run();
      	}
    2. Replace the annotation {Config} with your specific configurations.

    3. run ./run.sh under /ImageCodec/

  • Do the following commands on the terminal

    1. run chmod +x ./run.sh
    2. key in your pwd
    3. run ./run.sh
  • Use Matlab code to convert YUV to PNG

    • run genImage under /ImageCodec/ on Matlab terminal
  • Check the output files in /ImageCodec/results/

Encoding an image

  • This API gives 2 ways to encode an image.
  1. Specific upper bound of bitrate

    YUVImage* img = YUVImage::import({image_path})
    						->withFormat({image_format})
    						->withSize({image_width}, {image_height});
    PerformancePackage::load(img->getName());
    ImageCodec::encodeTo({target_path})->withImage(img)->withMaxBitrate({max_bitrate})->run();

    e.g.

    YUVImage* img = YUVImage::import("./1_1536x1024.yuv")
    						->withFormat(YUVImage::FORMAT_4_2_0)
    						->withSize(1536, 1024);
    PerformancePackage::load(img->getName());
    ImageCodec::encodeTo("./bitstream.bin")->withImage(img)->withMaxBitrate(0.75)->run();
    PerformancePackage::save();
  2. Specific lower bound of PSNR

    YUVImage* img = YUVImage::import({image_path})
    						->withFormat({image_format})
    						->withSize({image_width}, {image_height});
    PerformancePackage::load(img->getName());
    ImageCodec::encodeTo({target_path})->withImage(img)->withMinPSNR({min_PSNR})->run();
    PerformancePackage::save();

    e.g.

    YUVImage* img = YUVImage::import("./1_1536x1024.yuv")
    						->withFormat(YUVImage::FORMAT_4_2_0)
    						->withSize(1536, 1024);
    PerformancePackage::load(img->getName());
    ImageCodec::encodeTo("./bitstream.bin")->withImage(img)->withMinPSNR(28)->run();
    PerformancePackage::save();

Decoding an image

  • Decoding a compressed file is easy.

     ImageCodec::decode({bitstream_path})->saveTo({target_path})->run();

    e.g.

     ImageCodec::decode("./bitstream.bin")->saveTo("./decompressed.yuv")->run();

Obtaining PSNR

  • Importing 2 images and then call the following API.

     double PSNR = img1->calPSNR(img2);

    e.g.

     YUVImage* img1 = YUVImage::import("./1_1536x1024.yuv")
     						->withFormat(YUVImage::FORMAT_4_2_0)
     						->withSize(1536, 1024);
     YUVImage* img2 = YUVImage::import("./decompressed.yuv")
     						->withFormat(YUVImage::FORMAT_4_2_0)
     						->withSize(1536, 1024);
     double PSNR = img1->calPSNR(img2);

Obtaining Bitrate

  • Importing the original image, and then give the path of compressed file.

     double bitrate = img->calBitrate({bitstream_path});

    e.g.

     YUVImage* img = YUVImage::import("./1_1536x1024.yuv")
     						->withFormat(YUVImage::FORMAT_4_2_0)
     						->withSize(1536, 1024);
     double bitrate = img->calBitrate("./bitstream.bin");

Coding System

  • Level 1: main coding system.

level 1 coding

  • Level 2: optionally applied, for compensating PSNR.

level 2 coding

Work Flow

work flow

  1. Define the searching space. ( O(dsr x mk x pid x qres), very large! )
  2. Use all parameters to encode the file.
  3. Evaluate the bitrate and PSNR.
  4. Submit the result to ParameterPackage
  5. Find the best parameters set under specific constraint.

Bitstream Format (top-bottom, left-right)

coding format

Solution of slow parameters searching

To save the time re-searching same parameters at same image, an database/controller has been programmed for saving these parameters. It is /ImageCodec/libs/PerformancePackage.h:

  • the parameters will be saved in /ImageCodec/libs/PerformancePackageCache/, that's why this project should be runned at the appropriate directory. (to avoid some potential problems)

  • the database is programmed as singleton, which can be identified with 'name'. In this project, the parameters are stored in the database named from YUVImage::getName()

    i.e.

     YUVImage* image = YUVImage::import(mImgPath)
     								->withSize(mWidth, mHeight)
     								->withFormat(YUVImage::FORMAT_4_2_0);
     PerformancePackage::load(image->getName());

APIs List (only description and partial usage so far)

ImageCodec.h (642)

  • Main component of this project: to encode/decode a file.

    • Encoding

       ImageCodec::encodeTo({target_path})
       			->withImage({your_image})
       			->withMaxBitrate({max_bitrate})
       			->run();
       ImageCodec::encodeTo({target_path})
       			->withImage({your_image})
       			->withMinPSNR({min_PSNR})
       			->run();
    • Decoding

       ImageCodec::decode({bitstream_path})
       			->saveTo({decompressed_path})
       			->run();

YUVImage.h (520)

  • An object that contains lots of functions/properties that an image has.

    • Importing an image

       YUVImage* image = 
       		YUVImage::import({image_path})
       				->withSize({image_width}, {image_height})
       				->withFormat(YUVImage::FORMAT_4_2_0);
    • Initiate an empty image

       YUVImage* image = YUVImage::emptyImage({assigned_path}, 
       										{image_width}, 
       										{image_height}, 
       										{image_format});
    • Save the image into a file

       image->exportTo({tartget_path});
    • Properties

       image->getWidth(DataLayer::Y); image->getHeight(DataLayer::Cr);
       image->getDataSize(DataLayer::Cb);
       image->getDataSize(); // get total size of image
       image->getFormat(); // FORMAT_4_2_0 | FORMAT_4_2_2 | FORMAT_4_4_4
       image->getPath();
       image->getName();
    • Data getters & setters

       image->getYDataAt<int>(x, y);
       image->getCbDataAt<short>(x, y);
       image->getCrDataAt<long>(x, y);
       image->setYDataAt(x, y, (short)v);
       image->setCbDataAt(x, y, (short)v);
       image->setCrDataAt(x, y, (short)v);
    • Circular Addition/Subtraction

       YUVImage* image3 = image1->add(image2);
       YUVImage* image4 = image1->diff(image2);
    • Evaluate difference between two images

       double PSNR = image1->calPSNR(image2);
    • Evaluate the compressed bitrate

       double bitrate = image1->calBitrate({bitstream_path});

YUVImageFactory.h (150)

  • A toolbox to process YUVImage: to apply k-means clustering.

    • Clustering the image

       YUVImageFactory* factory = 
       			YUVImageFactory::initWithImage(image)
       								->useQuantization()
       								->withYUVLevels({Y_clusters}, {U_clusters}, {V_clusters})
       								->compress();
    • After clustering, all data in the image refer to cluster ID rather than magnitude. The clustered image with cluster ID can be obtained with YUVImage* symbolImg = factory->getSymbolImage(), and the clustered image with magnitude can be obtained with YUVImage* clusteredImg = factory->getImage()

    • Information after clustering

       factory->getYLevel(); // get # of Y_clusters
       factory->getYSymbolAt(n); // get the n^th center of Y
       factory->getULevel(); factory->getUSymbolAt(n);
       factory->getVLevel(); factory->getVSymbolAt(n);

ImagePredictor.h (139)

  • A toolbox to process YUVImage: to generate predicted residual image with a specific predictor ID.

    • Directly getting a residual image

       YUVImage* resImg = ImagePredictor::predictResidual(image, predictorID);
    • Use circular subtraction (e.g. 1 cirSubtr_5 7 = (1-7) % 5 = 4)

       int circularN[3] = {64, 16, 8};
       YUVImage* resImg = ImagePredictor::predictResidual(image, predictorID, circularN);
    • Predictor ID: (only 1-8 used)

      predictors

Transform.h (65)

  • A toolbox to do discrete cosine transform: 4x4-block supported so far. [not used]

    • Transformation of a vector

       int v1[4] = {1, -1, 2, -2};
       Transform::dct4<int>(v1);
       
       short v2[4] = {11, 1, -2, 1};
       Transform::idct4<short>(v2);
    • Transformation of multiple vectors

       int v1[12] = {1, 2, 1, 2, -1, 3, 15, 10, 0, 2, -3, -5};
       Transform::dct4<int>(v1, 3);
       
       long v2[8] = {1254, -4, 11, 19, 28, 3, 2, 0};
       Transform::idct4<long>(v2, 2);

KmeansFactory.h (176)

  • A toolbox to do k-means clustering.

    • KmeansFactory<T>: all data are stored in T type.
    • KmeansInfo<T, centerT>: all centers are stored in centerT type.
     int data[65536] = {1, 4, 2, 8, 10, 16, ... };
     KmeansFactory<int>* factory = KmeansFactory<int>::getInstance();
     KmeansInfo<int, int>* info = factory->withRawData(data, 65536)
     									->clusterInto(4096)
     									->run<int>();

HuffmanTable.h (298)

  • An object able to construct a canonical Huffman table with specific contents data.

    • HuffmanTable<T>: all data are stored in T type.
     char data[1024] = {0, 5, 123, ... };
     HuffmanTable<short>* table = HuffmanTable::initWithSize(256)
     											->withData(data, 1024);
    • Get maximal word length and entries value

       int WL_max = table->getMaxWordLength();
       for (int i = 0; i < WL_max; i++)
       	int nEntries = table->getNumEntriesWithWordLength(i+1);
    • Get symbol/probability at specific index

       Symbol<T>* symb = table->getSymbolAt(n);
       double p = table->getProbabilityAt(n);

GolombRiceTable.h (213)

  • An object able to construct a Golomb-Rice table. [not used]

    • GolombRiceTable<T>: all data are stored in T type
     GolombRiceTable<int>* table = GolombRiceTable<int>::initWithSize({size})
     													->withGroupSize({group_size})
     													->init();
    • Other getters are similar to HuffmanTable's.

PerformancePackage.h (316)

  • A central controller/database to manage the efficiency and its corresponded parameters set.

    • Getting database with a specific name

       PerformancePackage* pkg = PerformancePackage::getInstance(name);
    • Submitting a 'tried parameters set' to database

       CompressionParameters* 
       		cp = Builder()->withArea(imageArea)
       					  ->withDownSampleScale(wscale, hscale)
       					  ->withMk_YUV(mk_y, mk_u, mk_v)
       					  ->withPredictorID(predictorID)
       					  ->withKmeansTableSize(N_k)
       					  ->withMk_TABLE(mk_table)
       					  ->withMaxWordLength(WL_max)
       					  ->withKmeansTableBitrate(kmeansTableBitrate)
       					  ->withAdjustBitrate(adjustBitrate)
       					  ->withResidualQuantizationConst(Q_res)
       					  ->withPSNR(PSNR)
       					  ->withAdjust(isWithAdjust);
       pkg->submit(cp);
    • After searching all parameters, the database should eliminate some higher-bitrate and lower-PSNR (higher cost, lower efficiency) parameters set.

       pkg->anneal();
    • Save all data into the disk

       PerformancePackage::save();
    • Load data with a specific name

       PerformancePackage::load(name);

BitReader.h (93)

  • A toolbox to read a file in bits level.

    • Opening an input stream

       BitReader* reader = BitReader::open({stream_path});
    • Reading data (big endian)

       char v = reader->read(); // read a bit
       char* v = reader->read(size);
       long v = reader->read<long>(15); // read 15 bits and store in long
       short v = reader->read<short>(3);
    • Properties

       int restBits = reader->remains();
       int size = reader->getFileSize(); // in terms of bits not bytes
    • DO NOT FORGET TO CLOSE INPUT STREAM! (because it is singleton.)

       reader->close();

BitWriter.h (101)

  • A toolbox to write a file in bits level.

    • Opening the output stream

       BitWriter* writer = BitWriter::open({stream_path});
    • Writing data

       writer->write(0);
       
       char v[5] = {0, 1, 1, 1, 0};
       writer->write(v, 5);
       
       int k = 18;
       writer->write<int>(k, 7); // write '0010010'
    • DO NOT FORGET TO CLOSE OUTPUT STREAM! (because it is singleton.)

       writer->close();

Symbol.h (77)

  • An object that contains lots of functions/properties that a symbol has.

    • Symbol<T>: the value of the symbol is stored in T type

    • Getters

       Symbol<short>* s = new Symbol<short>(v);
       short value = s->getValue();
       std::vector<char> word = s->getWord();
       string wordStr = s->getWordString();
       int wordCount = s->getWordSize();

Total: 2790 lines

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published