Feature integration for image information retrieval using image mining techniques


Published on

1 Like
  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Feature integration for image information retrieval using image mining techniques

  1. 1. International Journal of Computer Engineering COMPUTER (IJCET), ISSN 0976 – INTERNATIONAL JOURNAL OF and Technology ENGINEERING 6367(Print), ISSN 0976 – 6375(Online) Volume 3, Issue 3, October-December (2012), © IAEME & TECHNOLOGY (IJCET)ISSN 0976 – 6367(Print)ISSN 0976 – 6375(Online)Volume 3, Issue 3, October - December (2012), pp. 273-281 IJCET© IAEME: www.iaeme.com/ijcet.aspJournal Impact Factor (2012): 3.9580 (Calculated by GISI) ©IAEMEwww.jifactor.com FEATURE INTEGRATION FOR IMAGE INFORMATION RETRIEVAL USING IMAGE MINING TECHNIQUES Madhubala Myneni Professor, Dept. of CSE , Mahaveer Institute of Science and Technology, Bandlaguda, vyasapuri, kesavagiri Post Hyderabad-5 baladandamudi@gmail.com Dr.M.Seetha Professor, Dept. of CSE, G.Narayanamma Institute of Technology & Science for Women, Shaikpet, Hyderabad smaddala2000@yahoomail.com ABSTRACT In this image information retrieval system the feature extraction process is analyzed and a new set of integrated features are proposed using image mining techniques. The content of an image can be expressed in terms of low level features as color, texture and shape. The primitive features are extracted and compared with data set by using various feature extraction algorithms like color histograms, wavelet decomposition and canny algorithms. This paper emphasizes on feature extraction algorithms and performance comparison among all algorithms. The system uses image mining techniques for converting low level semantic characteristics into high level Features, which are integrated in all combinations of primitive features. The feature integration has restricted to three different methodologies of feature Integration: color and texture, color and shape, texture and shape. It is ascertained that the performance is superior when the image retrieval based on the super set of integrated features, and better results than primitive set. Keywords-Feature Extraction, Feature Integration, Image Information Retrieval 1. INTRODUCTION Image Information Retrieval aims to provide an effective and efficient tool for managing large image databases. Image retrieval is one of the most exciting and fastest growing research areas in the field of medical imaging [2]. The goal of CBIR is to retrieve images from a database that are 273
  2. 2. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 –6367(Print), ISSN 0976 – 6375(Online) Volume 3, Issue 3, October-December (2012), © IAEMEsimilar to an image placed as a query. In CBIR, for each image in the database, features areextracted and compared to the features of the query image. A CBIR method typically converts animage into a feature vector representation and matches with the images in the database to findout the most similar images. In various studies different databases have been to compare thestudy.Content-based image retrieval (CBIR) has been an active research topic since 1990s. Suchsystems index images using their visual characteristics, such as color, texture and shape, whichcan be extracted from image itself automatically. They can accept an image example as query,and return a list of images that are similar to the query example in appearance. To extract theprimitive features of a query image and compare them to those of database images. The imagefeatures under consideration were colour, texture and shape. Thus, using matching andcomparison algorithms, the colour, texture and shape features of one image are compared andmatched to the corresponding features of another image. In the end, these metrics are performedone after another, so as to retrieve database images that are similar to the query. The similaritybetween features was to be calculated using algorithms used by well known CBIR systems suchas IBMs QBIC. For each specific feature there is a specific algorithm for extraction and anotherfor matching.The integration of structure features, which are particularly suitable for the retrieval of manmadeobjects, and color and texture features, which are geared towards the retrieval of natural imagesin general. Specifically, we restrict our attention to three different methodologies of featureIntegration: color and texture, color and shape, texture and shape results in better performancethan using shape, color, and texture individually. Retrieving the relevant images in imageretrieval system is a three step process: To extract the relevant primitive features of color, textureand shape, To design the filter for each feature, To generate the integrated features and performthe comparison.2. FEATURE EXTRACTIONFeature Extraction is the process of creating a representation, or a transformation from theoriginal data. The images have the primitive features like color, texture, shape, edge, shadows,temporal details etc. The features that were most promising were color, texture and shape/edge.The reasons are color can occur in limited range of set. Hence the picture elements can becompared to these spectra. Texture is defined as a neighborhood feature as a region or a block.The variation of each pixel with respect to its neighboring pixels defines texture. Hence thetextural details of similar regions can be compared with a texture template. shape/edge is simplya large change in frequency. The three feature descriptors mainly used most frequently duringfeature extraction are color, texture and shape.Colors are defined in three dimensional color spaces. These could either be RGB (Red, Green,and Blue), HSV (Hue, Saturation, and Value) or HSB (Hue, Saturation, and Brightness). The lasttwo are dependent on the human perception of hue, saturation, and brightness. Most imageformats such as JPEG, BMP, GIF, use the RGB color space to store information. The RGBcolor space is defined as a unit cube with red, green, and blue axes.Texture is that innate property of all surfaces that describes visual patterns, each havingproperties of homogeneity. It contains important information about the structural arrangement ofthe surface, such as; clouds, leaves, bricks, fabric, etc. It also describes the relationship of thesurface to the surrounding environment. In short, it is a feature that describes the distinctive 274
  3. 3. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 –6367(Print), ISSN 0976 – 6375(Online) Volume 3, Issue 3, October December (2012), © IAEME October-Decemberphysical composition of a surface.The most popular statistical representations of texture are: Co- surface.The Cooccurrence Matrix, Tamura Texture and Wavelet Transform.The classification of all methods among the two other approaches: structural and statistical.The statistical point of view, an image is a complicated pattern on which statistics can be heobtained to characterize these patterns. The techniques used within the family of statisticalapproaches make use of the intensity values of each pixel in an image, and apply various sstatistical formulae to the pixels in order to calculate feature descriptors. Texture featuredescriptors, extracted through the use of statistical methods, can be classified into two categories intoaccording to the order of the statistical function that is utilized: First-Order Texture Features and OrderSecond Order Texture Features. First Order Texture Features are extracted exclusively from theinformation provided by the intensity histograms, thus yield no information about the locations histograms,of the pixels. Another term used for FirstFirst-Order Texture Features is Grey Level DistributionMoments.Shape may be defined as the characteristic surface configuration of an object; an outline orcontour. It permits an object to be distinguished from its surroundings by its outline. Shape ur.representations are two categories Boundary Boundary-based and Region-based.Canny Edge detection is an optimal smoothing filter given the criteria of detection, localizationand minimizing multiple responses to a single edge. This method showed that the optimal filter ndgiven these assumptions is a sum of four exponential terms. It also showed that this filter can bewell approximated by first-order derivatives of Gaussians. Canny also introduced the notion of order Cannynon-maximum suppression, which means that given the pre smoothing filters, edge points are maximum pre-smoothingdefined as points where the gradient magnitude assumes a local maximum in the gradientdirection.Sobel edge detection operations are performed on the data and the processed data is sent back to thecomputer. The transfer of data is done using parallel port interface operating in bidirectional mode. Forestimating image gradients from the input image or a smoothed version of it, different gradient operators differentcan be applied. The simplest approach is to use central differences:corresponding to the application of the following filter masks to the image data:The well-known and earlier Sobel operator is based on the following filters: knownGiven such estimates of first- order derivatives, the gradient magnitude is then computed as: derivatives, 275
  4. 4. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 –6367(Print), ISSN 0976 – 6375(Online) Volume 3, Issue 3, October December (2012), © IAEME October-Decemberwhile the gradient orientation can be estimated as3. IMAGE INFORMATION RETRIEVALIn general all Image retrieval algorithms are based on Image Primitive features like color, texture ge coloand shape. In this paper, the proposal of Comb Combinational features is specified to give goodperformance. On each feature more efficient algorithms are used to retrieve the information fromdata set.3.1. Image Retrieval based on Color .1.In this paper color based image retrieval has performed in two steps. First for extracting color ps.feature information Global color histograms has used. To quantize the colors, number of bins are20. Second step is calculating the distance between bins by using Quadratic distance algorithm.The results are the distance from zero the less similar the images are in color similarity. m3.1.2 Color HistogramsGlobal color histograms extract the color features of images. To quantize the number of bins are20. This means that colors that are distinct yet similar are assigned to the same bin reducing the thenumber of bins from 256 to 20. This obviously decreases the information content of images, butdecreases the time in calculating the color distance between two histograms. On the other hand hekeeping the number of bins at 256 gives a more accurate result in terms of color distance. Lateron we went back to 256 bins due to some inconsistencies obtained in the color distances betweenimages. There hasnt been any evidence to show which color space generates the best retrievalresults, thus the use of this color space did not restrict us an anyway. hus3.1.3. Quadratic Distance AlgorithmThe equation we used in deriving the distance between two color histograms is the quadraticdistance metric: t ( d 2 ( Q , I ) = HQ − H I ) A( H Q − HI )The equation consists of three terms. The derivation of each of these terms will be explained inthe following sections. The first term consists of the difference between two color histograms; or colomore precisely the difference in the number of pixels in each bin. This term is obviously a vector obviouslysince it consists of one row. The number of columns in this vector is the number of bins in ahistogram. The third term is the transpose of that vector. The middle term is the similaritymatrix. The final result d represents the colo distance between two images. The closer the color tancedistance is to zero the closer the images are in color similarity. The further the distance from zerothe less similar the images are in color similarity. 276
  5. 5. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 –6367(Print), ISSN 0976 – 6375(Online) Volume 3, Issue 3, October-December (2012), © IAEME3.1.4. Similarity MatrixAs can be seen from the color histograms of two images Q and I, the color patterns observed inthe color bar are totally different. A simple distance metric involving the subtraction of thenumber of pixels in the 1st bin of one histogram from the 1st bin of another histogram and so on isnot adequate. This metric is referred to as a Minkowski-Form Distance Metric, which onlycompares the “same bins between color histograms “.This is the main reason for using the quadratic distance metric. More precisely it is the middleterm of the equation or similarity matrix A that helps us overcome the problem of different colormaps. The similarity matrix is obtained through a complex algorithm: 1  2 2 2 2 (  v − vi  q ) ( ( ) + sq cos hq − si cos(hi ) ) + (s q ( ) sin hq − si sin(hi ) )  aq ,i = 1 − 5Which basically compares one colour bin of HQ with all those of HI to try and find out whichcolor bin is the most similar? This is continued until we have compared all the color bins of HQ.In doing so we get an N x N matrix, N representing the number of bins. What indicates whetherthe color patterns of two histograms are similar is the diagonal of the matrix, shown below. If thediagonal entirely consists of one’s then the color patterns are identical. The further the numbersin the diagonal are from one, the less similar the color patterns are. Thus the problem ofcomparing totally unrelated bins is solved.3.2. Image Retrieval based on TextureIn this paper texture based image retrieval has performed in three steps. First for extractingstatistical texture information Pyramid-structured wavelet transform has used. Thisdecomposition has been done in five levels. Second step is calculating energy levels on eachdecomposition level. Third step is calculating Euclidian distance between query image anddatabase images. The top most five images from the list are displayed as query result.3.2.1. Pyramid-Structured Wavelet TransformThis transformation technique is suitable for signals consisting of components with informationconcentrated in lower frequency channels. Due to the innate image properties that allows formost information to exist in lower sub-bands, the pyramid-structured wavelet transform is highlysufficient.Using the pyramid-structured wavelet transform, the texture image is decomposed into four subimages, in low-low, low-high, high-low and high-high sub-bands. At this point, the energy levelof each sub-band is calculated .This is first level decomposition. Using the low-low sub-band forfurther decomposition, we reached fifth level decomposition, for our project. The reason for thisis the basic assumption that the energy of an image is concentrated in the low-low band. For thisreason the wavelet function used is the Daubechies wavelet.For this reason, it is mostly suitable for signals consisting of components with informationconcentrated in lower frequency channels. Due to the innate image properties that allows formost information to exist in lower sub-bands, the pyramid-structured wavelet transform is highlysufficient.3.2.2. Energy LevelEnergy Level Algorithm: • Decompose the image into four sub-images • Calculate the energy of all decomposed images at the same scale, using [2]: 277
  6. 6. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 –6367(Print), ISSN 0976 – 6375(Online) Volume 3, Issue 3, October-December (2012), © IAEME m n 1 E = MN ∑∑ X (i , j ) i =1 j =1 where M and N are the dimensions of the image, and X is the intensity of the pixel located at row i and column j in the image map. • Repeat from step 1 for the low-low sub-band image, until index ind is equal to 5. Increment ind.Using the above algorithm, the energy levels of the sub-bands were calculated, and furtherdecomposition of the low-low sub-band image. This is repeated five times, to reach fifth leveldecomposition. These energy level values are stored to be used in the Euclidean distancealgorithm.3.2.3. Euclidean DistanceEuclidean Distance Algorithm: - Decompose query image. - Get the energies of the first dominant k channels. - For image i in the database obtain the k energies. - Calculate the Euclidean distance between the two sets of energies, using [2]: k ∑ (x 2 Di = k =1 k − yi , k ) - Increment i. Repeat from step 3.Using the above algorithm, the query image is searched for in the image database. The Euclideandistance is calculated between the query image and every image in the database. This process isrepeated until all the images in the database have been compared with the query image. Uponcompletion of the Euclidean distance algorithm, we have an array of Euclidean distances, whichis then sorted. The five topmost images are then displayed as a result of the texture search.3.3. Image Retrieval based on ShapeIn this paper shape based image retrieval has performed in two steps. First for extracting edgefeature canny edge detection algorithm and sobel edge detection algorithm are used. Second stepis calculating Euclidian distance between query image and database images. The top most fiveimages from the list are displayed as query result.3.3.1. Canny Edge Detection AlgorithmThe Canny edge detection was developed by John F. Canny in 1986 and uses a multi-stagealgorithm to detect a wide range of edges in images. Stages of the Canny algorithm - Noise reduction - Finding the intensity gradient of the image - Non-maximum suppression - Tracing edges through the image and hysteresis threshold - Differential geometric formulation of the Canny edge detector - 278
  7. 7. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 –6367(Print), ISSN 0976 – 6375(Online) Volume 3, Issue 3, October-December (2012), © IAEME3.3.2. Sobel Edge detection:The implementation of image edge detection on hardware is important for increasing processingspeed of image processing systems. Edge information for a particular pixel is obtained byexploring the brightness of pixels in the neighborhood of that pixel. Measuring the relativebrightness of pixels in a neighborhood is mathematically analogous to calculating the derivativeof brightness. The pixel information extracted is transferred from the computer to the fieldprogrammable gate array device. sobel edge detection operations are performed on the data andthe processed data is sent back to the computer. The transfer of data is done using parallel portinterface operating in bidirectional mode.The algorithm mentioned in the previous section, for finding the Euclidean Distance betweentwo images is used to compute the similarity between the images. 4. Feature Integration All the extracted features are integrated to get the final extracted image as result. Every block has a similarity measure for each of the features. Hence after the feature extraction process each instance (block) is a sequence of 1s (YES) and 0s (NO) of length equal to the number of features extracted. Combining these extracted features is synonymous to forming rules. One rule that combines the three features is color & edge | textures, which means color AND edge OR texture. Depending on the features used and the domain, the rules vary. If a particular feature is very accurate for a domain, then the rule will assign the class label as YES (1) (1 in the table on the left). For those instances when I Class is not certain the class label is 2. This denotes uncertain regions that may or may not be existed. The same rule used during the training phase is also used in the testing phase. Table1: Rules for Feature Integration Rule Name Color Texture Edge Class-1 1 0 1 Class-2 0 0 0 Class-3 1 1 0 Class-4 1 0 0If there are 3 features, for example, the above table shows a part of a set of rules that could beused. The first and third rules say that color along with texture or edge conclusively determinesthat query image is present in that block. The second rule says that when none of the features is1 then query image is absent for sure. Fourth rule states that color on its own is uncertain indetermining the presence of query image. 279
  8. 8. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 –6367(Print), ISSN 0976 – 6375(Online) Volume 3, Issue 3, October-December (2012), © IAEME5. RESULTS AND OBSERVATIONSThe image database has used to retrieve the relevant images based on query image. The testimage database contains 200 images on various categories. Image retrieval has performed oncombinational feature set of primitive features like color, texture and shape.Based on commonly used performance measures in information retrieval, two statisticalmeasures were computed to assess system performance namely Recall and Precision. For goodretrieval system ideal values for recall is 1 and precision is of low value.The following table will give the performance evaluation of Image retrieval based on proposedIntegrated feature of color and Texture,Integrated Feature of Color and shape,Texture and shape. Table 2. comparitive analysis on three categories of integrated features Qery image categories Color and Texture Color and shape Texture and Shape Precision Recall Precision Recall Precision Recall Glass 0.444 1 0.833 1 1 0.8 Mobile 0.666 0.857 0.833 0.714 0.75 0.428 Apple 0.555 1 0.833 0.8 0.75 0.6 Narcissi Lilly 0.667 1 1 1 1 0.667 Horse 0.667 1 0.833 0.833 1 0.667 Cow 0.444 0.8 0.667 0.8 1 0.8 SunFlower 0.667 0.75 0.83 0.625 0.75 0.3756. CONCLUSION AND FUTURE ENHANCEMENTSThis paper elucidates the potentials of extraction of features of the image using color, texture andshape for retrieving the images from the specific image databases. The images are retrieved fromthe given database of images by giving the query image using color, texture and shape features.These results are based on various digital images of dataset. The performance of the imageretrieval was assessed using the parameters recall rate and precision. It was ascertained that therecall rate and precision are high when the image retrieval was based on the Feature Integrationon all the three features the color, texture and shape than primitive features alone. This work canbe extended further on huge data bases for retrieving relevant image. This can be extended forpixel clustering to obtain objects using different combination of weight for color and texture andshape features. It can maximize the performance by choosing the best combination between theseweights.REFERENCES 1. Aigrain, P et al (1996) “Content-based representation and ret rieval of visual media – a state-of-the-art review” Multimedia Tools and Applications 280
  9. 9. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976 –6367(Print), ISSN 0976 – 6375(Online) Volume 3, Issue 3, October-December (2012), © IAEME 2. Alsuth, P et al (1998) “On video retrieval: content analysis by Image Miner” in Storage and Retrieval for Image and Video Databases 3. Androutsas, D et al (1998) “Image retrieval using directional detail histograms” in Storage and Retrieval for Image and Video Databases 4. Bjarnestam, A (1998) “Description of an image retrieval system”, presented at The Challenge of Image Retrieval 5. Carson, C., Thomas, M., Belongie, S., Hellerstein, J. M., Malik, J., (1999), “Blobworld: A system for region-based image indexing and retrieval,” Third International Conference on Visual Information Systems, Springer. 6. Castelli, V. and Bergman, L. D., (2002), “Image Databases: Search and Retrieval of Digital Imagery”, John Wiley & Sons, Inc. 7. Craig Nevill-Manning (Google) and Tim Mayer (Fast). Anatomy of a Search Engine, Search Engine Strategies Conference, 2002 8. Daubechies, I., (1992), “Ten Lectures on Wavelets,” Capital City Press, Montpelier, Vermon.t 9. Drimbarean A. and Whelan P.F. (2001) Experiments in color texture analysis, P attern Recognition Letters, 22: 1161-1167. 10. Flickner, M., Sawhney, H., Niblack, W., Ashley, J., Huang, Q., Dom, B., Gorkani, M., Hafner, J., Lee, D., Petkovic, D., Steele, D. and Yanker, P., (1995), “Query by image and video content: The QBIC system,” IEEE Computer. 11. Hermes, T et al (1995) “Image retrieval for information systems” in Storage and Retrieval for Image and Video Database 12. J. Smith and S. Chang. VisualSEEK: A fully automated content-based image query system. ACM Multimedia: 87-98, November 1996. 13. Jain, R (1993) “Workshop report: NSF Workshop on Visual Information Management Systems” in Storage and Retrieval for Image and Video Databases 14. John R. Smith and Shih-Fu Chang. VisualSEEk: a fully automated content-based image query system, ACM Multimedia 96, Boston, 1996 15. Petkovic, D (1996) “Query by Image Content”, presented at Storage and Retrieval for Image and Video Databases 16. Sergey Brin and Lawrence Page (Google Founders). The Anatomy of a Large-Scale Hypertextual Web Search Engine, 7th International WWW Conference (WWW 98). Brisbane, Australia, 1998. 17. Shou-Bin Dong and Yi-Ming Yang. Hierarchical Web Image Classification by Multi- level Features, Proceedings of the First International Conference on Machine Learning and Cybernetics, Beijing, 2002 18. Smith, J., (2001), “Color for Image Retrieval”, Image Databases: Search and Retrieval of Digital Imagery, John Wiley & Sons, New York. 19. Unser, M., (1995), “Texture classification and Segmentation Using Wavelet Frames,” IEEE Transaction Image Processing 20. W. Ma, Y. Deng and B. S. Manjunath. Tools for texture/color based search of images.SPIE International conference - Human Vision and Electronic Imaging: 496-507,February 1997. 21. W. Niblack, R. Barber, and et al. The QBIC project: Querying images by content using color, texture and shape. In Proc. SPIE Storage and Retrieval for Image and Video Databases, Feb 1994. 281