ROI Based Image Compression in Baseline JPEG

M M M Kumar Varma Int. Journal of Engineering Research and Applications www.ijera.com
ISSN : 2248-9622, Vol. 4, Issue 9( Version 1), September 2014, pp.168-173
www.ijera.com 168 | P a g e
ROI Based Image Compression in Baseline JPEG
M M M Kumar Varma#1, Madhuri. Bagadi#2 Associate professor1, M.Tech Student2 Sri Sivani College of Engineering, Department of Computer Science Srikakulam, Andhrapradesh,, India ABSTRACT To improve the efficiency of standard JPEG compression algorithm an adaptive quantization technique based on the support for region of interest of compression is introduced. Since this is a lossy compression technique the less important bits are discarded and are not restored back during decompression. Adaptive quantization is carried out by applying two different quantization to the picture provided by the user. The user can select any part of the image and enter the required quality for compression. If according to the user the subject is more important than the background then more quality is provided to the subject than the background and vice- versa. Adaptive quantization in baseline sequential JPEG is carried out by applying Forward Discrete Cosine Transform (FDCT), two different quantization provided by the user for compression, thereby achieving region of interest compression and Inverse Discrete Cosine Transform (IDCT) for decompression. This technique makes sure that the memory is used efficiently. Moreover we have specifically designed this for identifying defects in the leather samples clearly.
I. INTRODUCTION
Images are being used very extensively to depict, analyze and share information in the world today. With the advent of new age gadgets like Smartphones and social networking websites like Facebook the use of images has increased even more. So, with the ever increasing usage of images, the need for efficient compression techniques without the loss of vital information is necessary. JPEG (Joint Photographic Experts Group) is a standardized compression method for organizing and storing digital images [1]. JPEG uses a lossy form of compression based on the discrete cosine transform [2]. There has been always a selectable tradeoff between storage size and image quality using JPEG. In Baseline Sequential Quantization the whole image can be compressed with desired quality,But the Region of Interest compression is not achieved. Adaptive Quantization should be used to obtain the desired compression for the subject and the background separately. Kakarala and Bagadi described that the traditional JPEG compression standard, unlike JPEG2000, does not prescribe a means for spatially-variable quantization of transform coefficients. This prevents useful features such as region-of-interest (ROI) coding, where, for example, the subject in a photograph may be lightly compressed while the background is heavily compressed to hide irrelevant details [3].
II. LITERATURE REVIEW
The American Mathematical Society described the process of Baseline compression algorithm (lossy compression) taking an example of image whose dimensions are 250x375 giving a total of 93750 pixels requiring 281250 bytes of storage. However, that image is finally compressed such that it takes 32414 bytes for storage i.e., the image has been compressed by a factor of roughly nine. This article describes how the image can be represented in such a small file (compressed) and how it may be reconstructed (decompressed) from this file. The color space transformation used in preparing the image for compression is explained in detail [5]. Aruna and Ramesh demonstrated a simple method for signaling adaptive quantization to the decoder, using the empty slots in standard baseline Huffman table. This is useful for ROI coding, which has the advantage of improving image quality in selected regions. They also showed how the encoder may be modified to produce an output that is in compliant with the standard and how the decoder may be modified to correctly recover the adaptively quantized image [6].The basic concepts and definitions involved in simple JPEG compression are referred. Description of differences between lossless and lossy compressions, Baseline sequential and progressive formats etc. Overview of the Baseline sequential JPEG encoding and decoding processes as a whole using flowcharts and various other figures [2]. Kakarala and Bagadi showed how the quantization may be adapted in each block and how the adaption may be signaled to the decoder in a memory efficient manner. They also explained how this adaptation allows ROI (Region of Interest) coding of subject and background. Examples are shown to illustrate adaptive quantization, signaling and decoding [3].
Ken cabeen and Peter gent explained the need for image compression and also illustrated the
RESEARCH ARTICLE OPEN ACCESS

ISSN : 2248-9622, Vol. 4, Issue 9( Version 1), September 2014, pp.
various equations of the DCT and its uses with image
compression. The JPEG method is described step by
step in detail. They also showed how the DCT
(Discrete Cosine Transformation) and the IDCT
(Inverse DCT) are applied to each block. This whole
process is demonstrated taking a matrix as an
example which represents the block of image-pixel
values [4].Tutorial-Po-Hong Wu, Ting-Yu Chen
describes the popular JPEG still image coding
format. The purpose is to compress images while
maintaining acceptable image quality. Also the focus
is on the DCT and quantization. The entropy coding
is also briefly discussed [7].
III. PROPOSED ALGORITHM
IV. DETAILED ALGORITHM OF
COMPRESSION
STEP 1: The image is broken into 8x8 Blocks of
Pixels.
STEP 2: Working from left to right,top to bottom
DCT(DiscreteCosine Transform) is applied to each
block.
STEP 3: Each block is compressed through
Quantization.
STEP 4: The array of compressed blocks that
constitute the image is stored in drastically reduced
amount of space.
STEP 5: When desired the image is re-constructed
through Decompression, a process that Uses Inverse
DCT.
STEPS 1-5 above use DCT based image
Compression[8].
V. HOW JPEG ALGORITHM WORKS
FOR A GIVEN IMAGE
First, the image is broken into 8x8 blocks of
pixels as shown in the image below.
Fig: Shows an image with 8x8 blocks and inset image
[image source[5]].
Now we convert this image into a matrix
representation of pixels. Let each pixel is represented
by p bits, the pixel value is in the range of (0, 2^p-1).
We represent this matrix as Original as shown below.
Fig: Matrix representation of the pixels [4]
Pixel values are shifted from (0,2^p-1) to (-2^p-
1,(2^p-1)-1).Since our p=8 so shift the values
from(0,255) to (-12,127) because DCT requires range
be centered around 0.Each value of the matrix from
left to right and top to bottom is leveled off by
subtracting 128.We represent the matrix by M.
Load the input
image
Identify the ROI(Region
Of Interest)
Compress the ROI and the rest of
the image with different Quality
values
Display the Resultant image

Fig: Levelled of matrix [4]
1) Discrete Cosine Transform (DCT):
DCT converts intensity function into weighted sum of periodic basis (cosine) functions. It identifies band of spectral information which can be thrown away without loss of quality .So DCT converts spatial domain to frequency domain. The DCT equation computes the i,jth entry of the DCT of an image
Where x,y range from 0 to N-1
To get the matrix form of the above equation, we will use the following equation
N is the size of the block. P(x, y) represents x yth element of the image represented by matrix P. DCT matrix of an 8x8 blocks is as shown below:
Fig: DCT matrix for 8x8 blocks [4] Resultant Matrix D is obtained by multiplying original matrix M with DCT matrix T and then again multiplying with transpose of DCT matrix T’ so that the original value of M is retained. The Discrete Cosine Transformation is accomplished by matrix multiplication
D=TMT’
Resultant Matrix D is obtained:
Fig: Resultant Matrix D. [4]
2) QUANTIZATION
Varying levels in 8x8 blocks of DCT coefficient in which compression and quantization are obtained by selecting specific quantization matrices are compressed by quantization.
Fig: Quantized 8x8 block of DCT coefficients (level 50). [4]

For a quality level greater than 50 (less compression, higher image quality), the standard quantization matrix is multiplied by (100-quality level)/50. For a quality level less than 50 (more compression, lower image quality), the standard quantization matrix is multiplied by 50/quality level. [4]
Fig: Quantized 8x8 block of DCT coefficients (level 10). [4]
Fig: Quantized 8x8 block of DCT coefficients (level 90). [4] Quantization is achieved by dividing each element in the transformed image matrix D by the corresponding element in the quantization matrix, and then rounding to the nearest integer value.
Quantization matrix Q50 is used:
Fig: Quantized Matrix [4]. In the matrix C obtained above, the lowest frequency coefficients situated in the upper –left corner are sensitive to human eye, so they are not discarded where as high frequency coefficients which are less important situated in the lower right corner are discarded leading to a lossy compression.
3) CODING
The image which is constituted of an array of compressed blocks is stored in a drastically reduced amount of space. An encoder converts all coefficients of quantized matrix C to a stream of binary data after which most of the coefficients are zeros.
Fig: zig-zag sequence of compressed binary data [4].
4) DECOMPRESSION
Then the image is re-constructed through decompression, a process that uses the Inverse DCT.Reconstruction of our image begins by decoding the bit stream representing the quantized matrix C.
퐶푖,푗=푟표푢푛푑( 퐷푖,푗 푄푖,푗 )
푅푖,푗= 푄푖,푗 ∗(퐶푖,푗)

5) INVERSE DISCRETE COSINE
TRANSFORM(IDCT)
The Inverse Discrete Cosine Transform (IDCT)
is applied to matrix R by multiplying transpose of
DCT matrix first T’ and then with DCT matrix T,
which is rounded to the nearest integer. Original
Matrix M is obtained by adding 128 to each entry so
that the pixel value range from (0, 2^p-1).
Fig: Output Matrix (Decompressed) [4].
VI. EXPERIMENTAL RESULTS
The process of coding was carried out with the
help of MATLAB. First with the quality value
provided by the user, the pure image is compressed
with the desired quality.Finally the selected portion
which is important to the user is compressed by
providing more quality and the portion of image
which is irrelevant to the user is compressed with less
quality, thus gaining Region of Interest compression.
This process is implemented on defective leather
samples. Here, only the defective part of the leather
sample is important and thus we can use Region of
interest compression by compressing the defect with
high quality so that the defect is very clear. Whereas
the rest of the sample can be compressed more as it
contains the irrelevant information. Thus, Region of
interest compression gives more efficient results(by
highlighting the defects) than compressing the whole
image with the same quality.
The original image of a leather sample is
Complete image compression with quality 95%
The size of this image which is compressed with
the same quality throughout is 35.5MB.
VII. REGION OF INTEREST
COMPRESSION
The Region of Interest is selected i.e. the defect
is selected by the user
restored image from dct

Quality of compression that is applied to the
background is 1% Quality of compression that is
applied to the subject(Defect) is 95%
The size of this image which is compressed based on
Region of Interest is 5.64MB.
VIII. CONCLUSION
In this paper, we describe a much efficient
method of compressing image based on the idea of
what is required and what is not in the image data,
and there by applying the required amount of
compression at different parts of the image. The user
can select any part of image and can enter the
required quality for compression, if the subject is
more important then more quality to the subject is
provided by the user and less to the background.
otherwise if the subject is irrelevant to user and
background is important then more quality is
provided to the background and less to the subject.
By using this we can create images that are memory-efficient
that contains only the required data from the
JPEG file.
REFERENCES
[1] http://en.wikipedia.org/wiki/Image_file_for
mats#JPEG.2FJFIF; Wikipedia
[2] http://en.wikipedia.org/wiki/JPEG;Wikipedi
a
[3] A method for signalling block-adaptive
quantization in baseline sequential JPEG
Kakarala, R. ; Bagadi, R. TENCON 2009 -
2009 IEEE Region 10 Conference;
10.1109/TENCON.2009.5396160
[4] Image Compression and Discrete Cosine
Transform ;Ken Cabeen and Peter Gent,
Math 45,College of Redwoods.
[5] http://www.ams.org/samplings/feature-colu
mn/fcarc-image-compression
[6] http://www.ijmer.com/papers/Vol2_Issue4/F
N2428292831.pdf
[7] Advance Digital Signal Processing-JPEG
for Still Image Compression-Tutorial-Po-
Hong Wu, Ting-Yu Chen
[8] Gonzalez, R.C. and R.E.Woods, “Digital
image processing”, New Jersey: Prentice
Hall, 2002
overall image

ROI Based Image Compression in Baseline JPEG

More Related Content

What's hot (14)

Viewers also liked (20)

Similar to ROI Based Image Compression in Baseline JPEG (20)

Recently uploaded (20)

ROI Based Image Compression in Baseline JPEG