Upload
others
View
13
Download
0
Embed Size (px)
Citation preview
Content Based image Retrieval using Error Diffusion Block
Truncation coding Features with Relevance Feedback
Mr.Vipul Mahajan#1, Mrs.Alka Khade*2, #Computer Science Department, Computer Science Department, Mumbai University, Mumbai University
Abstract— A new concept to index colour images using the features extracted from the error diffusion Block truncation coding (EDBTC). The
EDBTC produces two colour quantizes and a bitmap Image, which is further, managed using vector quantization (VQ) to create the image
feature Descriptor. Herein two features are presented namely, colour histogram feature (CHF),bit Pattern histogram feature (BHF) to measure
the similarity between a query image and the Target image in database. The CHF and BHF are calculated from the VQ-indexed colour quantized
and VQ- indexed bitmap image, respectively. The distance calculated from CHF and BHF can be utilized to measure the similarity between two
images. User relevance feedback is used to filter result and gives more specific result
Keyword- Error diffusionBlock Truncation coding, Colour Histogram Feature, Bit Pattern Histogram Feature, Vector Quantization
I. INTRODUCTION
Content-based image retrieval (CBIR), also known as query by image content (QBIC) “Content-based” means that the
search examines the contents of the image comparatively than the metadata such as keywords, tags, or descriptions related with
the image. In this situation might refer to colours, shapes, textures, or any other information that can be derived from the image
itself. CBIR is needed because searches that depend on purely on metadata are dependent on explanation quality and completeness.
Having humans manually explain images by entering keywords or metadata in a large database can be time consuming and may
not capture the keywords wanted to describe the image.
These image retrieval systems involve two phases, indexing and searching, to retrieve a set of similar images from the database,
the indexing phase extracts the image features from all of the images in the database which is advanced stored in database as
feature vector. In the searching phase, the retrieval system originates the image features from an image submitted by a user (as
query image), which are later utilized for performing similarity matching on the feature vectors stored in the data- base. The image
retrieval system finally proceeds a set of images to the user with a specific similarity criterion, such as color similarity and texture
similarity. Error diffusion is a type of half toning in which the quantization residual is distributed to Neighbouring pixels that have
not yet been processed. Its main use is to convert a multi-level image into a binary image.
Block Truncation Coding, or BTC,[22] is a type of lousy image compression technique for greyscale images. It divides
the original images into blocks and then uses a quantiser to reduce the number of grey levels in each block at the same time as
maintaining the same mean and
Standard deviation. A pixel image is divided into blocks of classically 4x4 pixels. For each block the Mean and Standard
Deviation of the pixel values are calculated; these statistics generally change from block to block. The pixel values selected for
each restored, or new, block are chosen so that each block of the BTC compressed image will have (approximately) the same
mean and standard deviation as the corresponding block of the original image.
Many previous systems have been developed to improve the retrieval accuracy in the content-based image
retrieval (CBIR) system. One type of them is to pay image features derived from the compressed data stream [1]–[4].
G. Qiu, “Color image indexing using BTC,”2003,[1] In this Block Truncation Coding is done by using Color Image indexing. It
is classical method and Decoding procedure is absent.
M. R. Gahroudi and M. R. Sarshar, “Image retrieval based on texture and color method in BTC-VQ
compressed domain,”2007,[2] in this Image Retrieval is base on texture as well as color method,here BTC-VQ domain is used. In
this also classical method is used and blocking effect and Contour false problem is present.
F.-X. Yu, H. Luo, and Z.-M. Lu, “Colour image retrieval using pattern co-occurrence matrices based on BTC and VQ,”
2011,[3[ in this by using pattern co-occurrence matrices, color image get retrieved which is based on BTC and VQ. In this Feature
extraction of different segment of image become difficult.
S. Silakari, M. Motwani, and M. Maheshwari, “Color image clustering using block truncation algorithm,”2009,[4] in this color
image clustering is used with BTC algorithm.In this feature separation become difficult
JASC: Journal of Applied Science and Computations
Volume V, Issue XII, December/2018
ISSN NO: 1076-5131
Page No:2162
As opposite to the classical method that extracts an image descriptor from the original image, this retrieval arrangement directly
makes image features from the compressed stream without first performing the decoding process. This type of retrieval
purposes to reduce the time computation for feature extraction/generation since most of the multimedia images are already
converted to compressed area before they are r ecorded in any s torage devices. In [1]–[4] the image features are directly
constructed from the typical block truncation coding (BTC) or halftoning-based BTC compressed data stream without
performing the decoding procedure.
The concept of the BTC [9] E. J. Delp and O. R. Mitchell, “Image compression using block truncation
coding,” 1979, is to look for a simpleset of characteristic vectors to replace the original images. Specifically, the BTC
compresses an image into a new area by dividing the original image into multiple nonoverlapped image blocks, and each block is
then represent with two extreme quantizers (i.e., high and low mean values) and bitmap image. Two subimages constructed by
the two quantizes and the consistent bitmap image are produced at the end of BTC encoding stage, which are later
communicated into the decoder module through the transmitter. To generate the bitmap image, the BTC scheme performs
thresholding operation using the mean value of each image block such that a pixel value greater than the mean value is regarded as
1 (white pixel) and vice versa.
Y.-G. Wu and S.-C. Tai, “An efficient BTC image compression tech-nique,”1998,[10] some attempts have been
addressed to improve the BTC reconstructed image quality and compression ratio, and also to reduce the time computation
Some applications have been proposed in the literature triggered by the successfulness of EDBTC, such as image
watermarking [11], [13], [14], inverse halftoning [16], data hiding [17], image security [18], and halftone classification [15]. The
EDBTC scheme performs well in those areas with capable results, as described in [11]–[15], since it provides better reconstructed
image quality than that of the BTC scheme. the concept of the EDBTC compression is provided to the CBIR area, in
which the image feature descriptor is constructed from the EDBTC compressed data stream.
The descriptor is directly derived from EDBTC color quantizes and bitmap image in compressed area by involving the
vector quantization (VQ) for the indexing. The similarity standard between the query and target images is simply measured
using the EDBTC feature descriptor.
The contribution can be summarised are as follows 1) Corel 2000 image database is used 2) User relevance feedback is used
.
II. PROPOSED SYSTEM
Fig 1 shows Architecture of Image Retrieval system There are two parts of the system offline and online respectively In
offline part Image database compressed by EDBTC. In EDBTC compression color quantization and bitmap image quantization
involved. Color quantization further computed by color histogram and Bitmap image quantization further compute by pattern
histogram respectively In offline part Image database compressed by EDBTC. In EDBTC compression color quantization and
bitmap image quantization involved. Color quantization further computed by color histogram and Bitmap image quantization
further compute by pattern histogram respectively In offline part feature extraction occurred CHF & BHF of particular image
stored in image database With CHF and BHF.In online part user gives image query that query image compressed by EDBTC
Features get extracted from EDBTC compression & similarity compression feature extracted from Image database Image Retrived
by matching similarity compression further by using Relevance feedback Final Image get Retrieved
Fig. . 1: System Architecture
JASC: Journal of Applied Science and Computations
Volume V, Issue XII, December/2018
ISSN NO: 1076-5131
Page No:2163
III. EDBTC PROCESSING for COLOR IMAGE
Fig. . 2: Processing for color image
A. The EDBTC compresses an image in an effective way by including the error diffusion kernel to generate a bitmap image.
Simultaneously, it produces two extreme quantizes, specifically, minimum and maximum quantizes. The EDBTC system have a
great advantage in its low computational complexity in the bitmap image and two extreme quantizes group
B. The EDBTC bitmap image can be obtained by performing thresholding of the interbank average value with the error kernel. In a
block based process, the raster-scan path (from left to right and top to bottom) is applied to process each pixel
C. The EDBTC performs the thresholding operation by including the error kernel. We first need to compute the minimum,
maximum, and mean value of the interband average pixels
D. The bitmap image is made
E. The intermediate value is also generated at the same time with the bitmap image generation.
F. The residual quantization error of EDBTC can be calculated
G. The EDBTC thresholding process is performed in a successive way. One pixel is only processed once, and the residual
quantization error is diffused and gathered in to the neighboring unprocessed pixels.
H. The two extreme quantizes consist of RGB color information obtained by searching the minimum and maximum of value in an
image block for each RGB color space. Two EDBTC color quantizes are computed by looking for the minimum and maximum of
all image pixels in each image block.
IV. METHODOLOGY
I. The EDBTC compresses an image in an effective way by including the error diffusion kernel to generate a bitmap
image. Simultaneously, it produces two extreme quantizes, specifically, minimum and maximum quantizes. The EDBTC system
have a great advantage in its low computational complexity in the bitmap image and two extreme quantizes group, here Floyd
steinberg error kernel is used.[22]
II. The EDBTC bitmap image can be obtained by performing thresholding of the interbank average value with the error
kernel. In a block based process, the raster-scan path (from left to right and top to bottom) is applied to process each pixel in
a given image. Suppose that f (x , y) and f̄ (x , y) denote the original and interband average value, respectively. The interband
average value can be calculated as
JASC: Journal of Applied Science and Computations
Volume V, Issue XII, December/2018
ISSN NO: 1076-5131
Page No:2164
The fR(x, y), fG(x, y), and fB(x, y) denote the image pixels in the red, green, and blue (RGB) color
channels, respectively. The interband average image can be viewed as the grayscale type of a color image.
III. The EDBTC performs the thresholding operation by including the error kernel. We first need to compute the minimum,
maximum, and mean value of the interband average pixels as
∀∀∀∀x,y
∀∀∀∀x,y
IV. The bitmap image h(x, y) is made using the following rule:
V. The intermediate value o(x, y) is also generated at the same time with the bitmap image generation. The value o(x, y) can
be calculated as
(6)
VI. The residual quantization error of EDBTC can be calculated as
(7)
VII. The EDBTC thresholding process is performed in a successive way. One pixel is only processed once, and the residual
quantization error is diffused and gathered in to the neighboring unprocessed pixels. The value f̄ (x , y) of unprocessed yet pixel is
updated using the following strategy:
(8)
Where is the error kernel to diffuse the quantization residual into its neighboring pixels which have not yet been
processed in the EDBTC thresholding
VIII. The two extreme quantizes consist of RGB color information obtained by searching the minimum and maximum of value
in an image block for each RGB color space. Two EDBTC color quantizes are computed by looking for the minimum and
maximum of all image pixels in each image block as
∀∀∀∀x,y ∀∀∀∀x,y ∀∀∀∀x,y
∀∀∀∀x,y ∀∀∀∀x,y ∀∀∀∀x,y
JASC: Journal of Applied Science and Computations
Volume V, Issue XII, December/2018
ISSN NO: 1076-5131
Page No:2165
A. Vector Quantization
VQ compresses an image in lossy method based on block coding principle. It is a fixed-to-fixed length algorithm. The
VQ finds a codebook by iteratively partitioning a given source vector with its known statistical properties to produce the code
vectors with the smallest average distortion when a distortion measurement is given as,
Let C = {c1 , c2 , . . . ,cNc } be the color codebook generated using VQ consisting Nc code words. The VQ needs many
images involved as the training set. The vector ck contains RGB color (or grayscale) information which is identical to the two
EDBTC quantizes. Except for generating the color codebook, it is also required to construct the bit pattern codebook
B=(B1,B2….BNb) consisting of Nb binary code words. For generating the bit pattern codebook, bitmap images are required as
input training set in the VQ process.
B. Color Histogram Feature (CHF)
The CHF is derived from the two EDBTC color quantizes, while BHF is computed from EDBTC bitmap image. In this
paper, the CHFmin and CHFmax are developed from the color minimum and maximum quantizers, respectively. The CHFmin
and CHFmax capture color information from a given image. These features represent the combination of pixel brightness and
color distribution in an image.
C. Bit Pattern Histogram Feature (BHF)
Another feature generated from a VQ-indexed EDBTC data stream is the BHF. This feature captures the visual pattern,
edge, and textural information in an image. The BHF can be obtained by tabulating the occurrence of a specific bit pattern
codebook in an image
D. Image Retrieval with EDBTC Feature
The similarity distance computation is needed to measure the similarity degree between two images. The distance
plays the most important role in the CBIR system since the retrieval result is very sensitive with the chosen distance metric. The
image matching between two images can be performed by calculating the distance between the query image given by a user
against the target images in the database based on their corresponding features (CHF and BHF). After the similarity distance
calculation, the system returns a set of retrieved image ordered in ascending manner based on their their similarity distance
scores .
/
+ /
/
E. Relevance feedback
In this paper user relevance feedback is used, after retrieving images user need to select correct output image from
retrieved images list so after sending selective images, it gives final relevant images
IV Performance Measurement
An effectiveness of the proposed EDBTC retrieval system is measured with the precision, recall and average precision and recall
values rate values. These values indicate the percentage of relevant image retrieved a CBIR system after user relevance feedback
with a specific number of retrieved images L. The precision (P(q)) and recall (R(q)) values are defined as
(P(q)) =nq / L (12)
(P(q)) =nq / Nq (13)
Where nq and Nq denotes the number of relevant image against a query image q,and the number of all relevant images against a
query image q in database
JASC: Journal of Applied Science and Computations
Volume V, Issue XII, December/2018
ISSN NO: 1076-5131
Page No:2166
Fig. 3 Image sample from Corel 2000
V.EXPERIMENTAL RESULT
In this section we test 10 different images on the basis of six different features like CHFmin, CHF max,BHF,CHF min
+CHF max,CHF min +BHF ,CHF max + BHF and CHF min +CHF max +BHF. We record correct retrieved images out of 30
images, precision and recall and time complexity of each image with respect to different features.
Fig. 4 Graph (a) to (j) Precision vs Recall
JASC: Journal of Applied Science and Computations
Volume V, Issue XII, December/2018
ISSN NO: 1076-5131
Page No:2167
Above all graph from (a) to (j) shows Precision vs Recall for different 10 images separately .following table shows
average values of recall. Precision and time for ten images with six features and figure 5 shows graph of average values of
precision vs recall.
Table I Average values of Recall, Precision and Time for ten images with six features
FEATURES NAME RECALL PRECISION TIME COMPLEXITY
CHFMIN+CHFMAX+BHF 0.11 0.38 1 MIN 9 SEC
CHFMAX+BHF 0.10 0.34 1 MIN 35 SEC
CHF MIN+BHF 0.09 0.31 1 MIN 17 SEC
CHFMIN+CHFMAX 0.10 0.34 1 MIN 48 SEC
BHF 0.10 0.34 1 MIN 58 SEC
CHFMAX 0.11 0.35 2 MIN 15 SEC
CHFMIN 0.09 0.32 2 MIN 13 SEC
Fig. 5 Average of ten images- Precision vs Recall
Fig. 6 Steps of system
Figure 6 shows different steps of system fig (a) shows browsing image from database ,(b) shows EDBTC compression-
intermediate image , (c) shows EDBTC compression -bitmap image with error, (d) shows bitmap image, minimum quantization
and maximum quantization ,(e) shows features extraction of bitmap image, minimum quantization and maximum quantization ,(f)
shows retrieved images ,(g) shows user selecting relevant images ,(h) shows final images with respect to query images .
JASC: Journal of Applied Science and Computations
Volume V, Issue XII, December/2018
ISSN NO: 1076-5131
Page No:2168
VI. CONCLUSION
In Content Based image Retrieval using Error Diffusion Block Truncation coding Features with Relevance feedback,
Error Diffusion improves image quality comparative to only Block truncation coding. we use the Relevance feedback at the last
step so it will reduce unnecessary image which are not match with input image. So, by using feature and Relevance feedback we
will try to improve accuracy of retrieval system as well as quality of image. We test six features with respect to time
complexity ,precision and recall for ten images and finally collect average values of all ten images so it gives good result for
average values
ACKNOWLEDGMENT
For this paper my guide,my friends and colegues helps me.
I acknowledge Mrs.Alka Khade,mrs.Soorya Nair and other contributors for giving proper idea and directionto for making this
paper
REFERENCES
[1] G. Qiu, “Color image indexing using BTC,” IEEE Trans. Image Process., vol. 12, no. 1, pp. 93–101, Jan. 2003.
[2] M. R. Gahroudi and M. R. Sarshar, “Image retrieval based on textureand color method in BTC-VQ compressed domain,” in Proc. Int.
Symp. Signal Process. Appl., Feb. 2007, pp.1-4.
[3] ]. F.-X. Yu, H. Luo, and Z.-M. Lu, “Colour image retrieval using pattern co-occurrence matrices based on BTC and VQ,” Electron. Lett.,
vol. 47,no. 2, pp. 100–101, Jan. 2011
[4] S. Silakari, M. Motwani, and M. Maheshwari, “Color image clustering using block truncation algorithm,” Int. J. Comput. Sci. Issues, vol.
4, no. 2, 2009, pp. 31–35.
[5] Z.-M. Lu and H. Burkhardt, “Colour image retrieval based on DCT-domain vector quantisation index histograms,” Electron. Lett., vol.
41, no. 17, pp. 956–957, 2005.
[6] P. Poursistani, H. Nezamabadi-Pour, R. A. Moghadam, and M. Saeed, “Image indexing and retrieval in JPEG compressed
domain based on vector quantization,” Math. Comput. Model., vol. 57, nos. 5–6, pp. 1005–1017, 2013. [Online]. Available:
http://dx.doi.org/10.1016/j.mcm.2011.11.064
[7] M. E. ElAlami, “A novel image retrieval model based on the most relevant features,” Knowl.-Based Syst.,
vol. 24, no. 1, pp. 23–32, 2011.
[8] J.-M. Guo, H. Prasetyo, and H.-S. Su, “Image indexing using the color and bit pattern feature fusion,” J. Vis.
Commun. Image Represent., vol. 24, no. 8, pp. 1360–1379, 2013.
[9] E. J. Delp and O. R. Mitchell, “Image compression using block truncation coding,” IEEE Trans.
Commun., vol. 27, no. 9, pp. 1335–1342, Sep. 1979.
[10] ] Y.-G. Wu and S.-C. Tai, “An efficient BTC image compression tech-nique,” IEEE Trans. Consum. Electron., vol. 44, no. 2, pp. 317–
325, May 1998.
[11] J.-M. Guo and Y.-F. Liu, “Joint compression/watermarking scheme using majority-parity guidance and halftoning-
based block truncation coding,”IEEE Trans. Image Process., vol. 19, no. 8, pp. 2056–2069, Aug. 2010.
[12] J.-M. Guo, “Improved block truncation coding using modified error diffusion,” Electron. Lett., vol. 44, no. 7,
Mar. 2008, pp. 462–464.
[13] J.-M. Guo, S.-C. Pei, and H. Lee, “Watermarking in halftone images with parity-matched error diffusion,”
Signal Process., vol. 91, no. 1, pp. 126–135, 2011
[14] S.-C. Pei and J.-M. Guo, “Hybrid pixel-based data hiding and block-based watermarking for error-diffused
halftone images,” IEEE Trans. Circuits Syst. Video Technol., vol. 13, no. 8, pp. 867–884, Aug. 2003.
[15] Y. F. Liu, J. M. Guo, and J. D. Lee, “Halftone image classification using LMS algorithm and naive Bayes,” IEEE
Trans. Image Process., vol. 20, no. 10, pp. 2837–2847, Oct. 2011
[16] Y.-F. Liu, J.-M. Guo, and J.-D. Lee, “Inverse halftoning based on the Bayesian theorem,” IEEE Trans.
Image Process., vol. 20, no. 4, pp. 1077–1084, Apr. 2011.
[17] J.-M. Guo and Y.-F. Liu, “High capacity data hiding for error-diffused block truncation coding,” IEEE Trans.
Image Process., vol. 21, no. 12,pp. 4808–4818, Dec. 2012.
[18] ] J.-M. Guo and Y.-F. Liu, “Halftone-image security improving using overall minimal-error searching,”
IEEE Trans. Image Process., vol. 20, no. 10, pp. 2800–2812, Oct. 2011.
[19] Color Image Clustering using Block Truncation Algorithm IJCSI International Journal of Computer Science
Issues, Vol. 4, No. 2, 2009
JASC: Journal of Applied Science and Computations
Volume V, Issue XII, December/2018
ISSN NO: 1076-5131
Page No:2169
[20] HSV Color Motif Co-Occurrence Matrix for Content based Image Retrieval International Journal of Computer
Applications (0975 – 888) Volume 48– No.16, June 2012
[21] Enhancement Image Compression Using BTC Algorithm Volume 4, Issue 2, February 2014 ISSN: 2277 128X
[22] Vipul R. Mahajan and Alka Khade “A survey;:Content Based Image Retrieval using Block Truncaton Coding” Vol 7,No. 12 December 2017
JASC: Journal of Applied Science and Computations
Volume V, Issue XII, December/2018
ISSN NO: 1076-5131
Page No:2170
JASC: Journal of Applied Science and Computations
Volume V, Issue XII, December/2018
ISSN NO: 1076-5131
Page No:2171