460 likes | 2.36k Views
On the Process of Realizing the Best BinDCT Configuration for Image Compression. Mahmoud AL-Ghreify Prof. D. Harvey Dr. C.W.Murphy Prof. D. Burton. Presentation outline. Motivation Introduction & Backgrounds Objectives of the research Results of Investigations Conclusions.
E N D
On the Process of Realizingthe Best BinDCT Configuration for Image Compression Mahmoud AL-Ghreify Prof. D. Harvey Dr. C.W.Murphy Prof. D. Burton
Presentation outline • Motivation • Introduction & Backgrounds • Objectives of the research • Results of Investigations • Conclusions
Motivation • Why image compression is important to us? • Medical applications • Multimedia based web applications • Wireless and video phones • The recent growth of image compression algorithms requires more efficient ways to encode signals and images
Introduction What is the Image compression? • Process to reduce the size of the image How ? • Using Discrete Cosine Transform Why DCT? • DCT can approximate linear signals well with few coefficients
Transform DCT Quantization F(u,v) f(I,j) Input Pixels Coding Tables Quantization Table DPCM Entropy Coding Data Output Zig Zag RLE
How DCT works? • When an image is processed by the DCT the image will be decomposed into low frequency and high frequency coefficients, this allows for further image compression operations to occur
Effect of the DCT transform Pixel values DC term of the horizontal basis functions is to the left, and the DC term for vertical basis functions is at the top. The top row and left column have 1-D intensity variations.
Lena Origin Image Lena 2-D DCT Image Energy compactness on the Image
JPEG DCT Versions • The floating version based scaled DCT algorithm with five floating multiplications and 29 additions for each dimension. • Fast integer version with five fixed point multiplications. • The slow integer version, a variation of the Loeffler’s algorithm with 12 fixed point multiplications and 32 additions. • the encoder of the selected H.263 is based on chen’s factorization with floating point multiplications.
Objectives • Construct a generic system, this modelshould be able to switch between the different BinDCT configurations to achieve better image compression. • Proposing pre-processing stage so the system can detect the best BinDCT configuration to be use with each incoming tile.
Selection-based techniques • RMSE(Relationship between each configuration Root Mean Square Error and the true DCT were obtained) • Entropy is a quantitative entity determines the amount of the information enclosed in a message. • Homogeneity is largely related to the local information extracted from an image and Indicate how uniform a region is
Investigations & Results • All nine configurations of the forward and Inverse BinDCT were implemented in C code. • Root Mean Square Error Calculations. • Entropy selection-based. • Homogeneity selection-based
Origin Image BinDCT1 BinDCT9 Forward BinDCT/InvBinDCT’s
Conclusions • Great performance can be achieved if the FPGA can dynamically switch between the different configurations of the BinDCT transforms • Two different pre-processing stage proposed to achieve this aim. • Entropy and Homogeneity selection-based prove to have the ability to detect best configuration for each incoming tile.