EC4091-DIGITAL SIGNAL PROCESSING LAB
ELECTRONICS AND COMMUNICATION ENGINEERING
NATIONAL INSTITUTE OF TECHNOLOGY, CALICUT
LAB PROJECT REPORT
Submitted by:
Ashna KK B160750EC
GSV Swetha B160900EC
Gali Amoolya B160619EC
Gopika Vijay B160513EC
JPEG IMAGE COMPRESSION IN MATLAB
ABSTRACT:
Original image generated by a camera sensor is a very large store, and
therefore is not efficient. The present scenario requires the use of
efficient image compression techniques to fulfill faster image transfers
and low storage space requirements. JPEG is often applied for the
processing and storage of full-color images with realistic elements and
brightness and color transitions.Also this format is used for storing and
transmitting graphical digital content (photos, scanned copies of digitized
pictures).
It is the most convenient for transmission of compressed images on the
Internet, because it takes less space in comparison to other formats.
However, JPEG is ideal for home photo storage, photo transmission
through the Internet or posting on the site.
INTRODUCTION:
JPEG is an image compression standard to store image in compressed
format. It represents the Joint Photographic Experts Group. Excellent
quality of JPEG is that it achieves high compression ratio and quality is
with almost no loss. JPEG format is very popular, and is used in a
large-sized image switching a plurality of devices such as digital
cameras, and is selected in the bandwidth-limited environments, such as
the format of the Internet. JPEG algorithm is best suited for photos and
realistic scenes with smooth changes in tone and color painting. JPEG is
not suitable for use with many edges and sharp changes, since this may
result in many image artifacts in the resulting image. In these cases, it is
best to use a lossless format such as PNG, TIFF or GIF. For this reason,
JPEG is not in use for medical and scientific applications, where the
image needs to be exact and slight error results into no reproduction of
captured data.
JPEG image may accept further losses, if it is frequently edited, and
then save it. The operation of decompression and recompression can
further reduce image quality. To solve this problem, the image should be
edited and saved in a lossless format, only converted to JPEG format,
just before the final transport to the required media. This ensures
minimal loss due to frequent savings. Saved as JPEG image files usually
have extensions such as .jpg, jpeg, or .jpeg .
BLOCK DIAGRAM:
THEORY AND IMPLEMENTATION:
Color Space Conversion:
On this step the colors of the RGB (Red, Green, Blue) image are
converted to YCbCr (Luminance, Chrominance Blue, Chrominance
Red). During this conversion we take three standard channels (RGB)
and convert them into a different representation which is based on a
luminance (brightness) channel and two opposing color channels. These
three channels are typically less interdependent then the RGB channels.
That allows us to store them with different resolution. If the image is
CMYK or grayscale, it is processed without any color conversion.
Chroma Subsampling:
It is widely known that we are much more sensitive to changes in
luminance than in chrominance. It means that we may neglect larger
changes in the chrominance without affecting our perception of the
image. Therefore, we can store the color information at lower resolution
than the luminance information. This method can be applied only to
images in YCbCr color space. CMYK images use all the channels to
store color information. Therefore each color channel is compressed and
quantized with the similar quality. Grayscale images have no color
information and do not need being converted.
Segmentation into Blocks:
The image is divided into 8 by 8 blocks of pixels. Since each block is
processed without reference to the others, further steps will be described
relatively to a single block.
Discrete Cosine Transform:
Each of the color component data (e.g. Y, Cb, Cr for YCbCr or C, M, Y,
K for CMYK) undergoes the Discrete Cosine Transform (DCT). The DCT
is similar to the Fourier Transform in the sense that it produces a kind of
a spatial frequency spectrum.
Quantization:
The amplitudes of the frequency components are quantized. The human
vision is much more sensitive to small variations in color or brightness
over large areas (low-frequency components) than to variations which
occur on every pixel (high-frequency components). Therefore, the
high-frequency components are stored with a lower accuracy than the
low-frequency ones. The quality setting of the encoder affects the
resolution of each frequency component. Higher values of quality setting
correspond to higher quality images and larger file sizes.
Huffman encoding:
Huffman coding is a method that takes symbols (e.g. bytes, DCT
coefficients, etc.) and encodes them with variable length codes that are
assigned according to statistical probabilities. A frequently-used symbol
will be encoded with a code that takes up only a couple bits, while
symbols that are rarely used are represented by symbols that take more
bits to encode. Here the resulting data for all 8 by 8 blocks of pixels is
compressed with a lossless algorithm, a variant of Huffman encoding.
Mean Square Error and Peak SNR:
Comparing restoration results requires a measure of image quality. Two
commonly used measures are Mean-Squared Error and Peak
Signal-to-Noise Ratio. The mean-squared error (MSE) between two
images g( x, y) and h(x,y) of size MXN is:
M N
M SE = 1
MN ∑ ∑ [h(n.m) − g (n, m)] 2
n=1 m=1
One problem with mean-squared error is that it depends strongly on the
image intensity scaling. A mean-squared error of 100.0 for an 8-bit
image (with pixel values in the range 0-255) looks dreadful; but a MSE of
100.0 for a 10-bit image (pixel values in [0,1023]) is barely noticeable.
Peak Signal-to-Noise Ratio (PSNR) avoids this problem by scaling the
MSE according to the image range.
P SN R =− 10log 10 MSSE
2
where S is the maximum pixel value. PSNR is measured in decibels
(dB). The PSNR measure is also not ideal, but is in common use. Its
main failing is that the signal strength is estimated as S 2 , rather than the
actual signal strength for the image. PSNR is a good measure for
comparing restoration results for the same image.
OBSERVATIONS:
Figure 1: Color components
Figure 2: Downsampling image by factor 10
Figure 3: DCT coefficients for one 8X8 block
Figure 4: Original input image before compression
Figure 5: Final output compressed image
Figure 6: Compression ratio, MSE and peakSNR obtained on
downsampling by two
Figure 7: MSE and PSNR on downsampling by a factor of 10
OBSERVATIONS AND INFERENCES:
Compression Ratio:
Initially the colour components are downsampled by a factor of 2. A
compression ratio of a 6.782 is obtained.
Upon downsampling by 5, the compression ratio decreased to 6.2515.
Further increasing downsampling rate to 10, the compression ratio
decreased further down to 5.7844.
So an inverse relation between compression ratio and downsampling
rate can be inferred from the above observations
Mean Square Error:
At a downsampling rate of 2, a MSE of 2.9439 is observed for YCr colour
component. When increased to 10, MSE increases to 5.5757. Similar
increase in MSE is observed for all colour components as given in figure.
PSNR:
PSNR value of 40.067 is obtained when downsampled by a factor of 2
and 38.067 when downsampled by a factor of 5. PSNR decreases by
increasing downsampling rate.
APPENDIX
Matlab code:
clc;
clear all;
close all;
I = imread('peppers.png');
Y_d = rgb2ycbcr( I );
% Downsample:
Y_d(:,:,2) = 2*round(Y_d(:,:,2)/2);
Y_d(:,:,3) = 2*round(Y_d(:,:,3)/2);
% DCT compress:
A = zeros(size(Y_d));
B = A;
Q = [16 11 10 16 24 40 51 61 ;
12 12 14 19 26 28 60 55 ;
14 13 16 24 40 57 69 56 ;
14 17 22 29 51 87 80 62 ;
18 22 37 56 68 109 103 77 ;
24 35 55 64 81 104 113 92 ;
49 64 78 87 103 121 120 101;
72 92 95 98 112 100 103 99];
for channel = 1:3
for j = 1:8:size(Y_d,1)-7
for k = 1:8:size(Y_d,2)
II = Y_d(j:j+7,k:k+7,channel);
freq = chebfun.dct(chebfun.dct(II).').';
freq = Q.*round(freq./Q);
A(j:j+7,k:k+7,channel) = freq;
B(j:j+7,k:k+7,channel) =
chebfun.idct(chebfun.idct(freq).').';
end
end
end
figure
imshow(ycbcr2rgb(Y_d))
title('Original')
figure
imshow(ycbcr2rgb(uint8(B)));
title('Compressed')
size(ycbcr2rgb(Y_d))/size(ycbcr2rgb(uint8(B)));
ImageSize = 8*prod(size(I))
CompressedImageSize = 8*nnz(A(:,:,1)) +
7*nnz(A(:,:,2)) + 7*nnz(A(:,:,3))
ImageSize/CompressedImageSize
/////huffmann coding
b = A(:);
b = b(:);
b(b==0)=[]; %remove zeros.
b = floor(255*(b-min(b))/(max(b)-min(b)));
symbols = unique(b);
prob = histcounts(b,length(symbols))/length(b);
dict = huffmandict(symbols, prob);
enco = huffmanenco(b, dict);
FinalCompressedImage = length(enco)
ImageSize/length(enco)
mean_square_error=MSE(ycbcr2rgb(Y_d),ycbcr2rgb(uint8
(B)))
peaksnr = psnr(ycbcr2rgb(Y_d),ycbcr2rgb(uint8(B)))
