36
Lov, Ashish, Shikha, Prateek CSE’09 Guru Gobind Singh Indraprastha University

Visual Search

Embed Size (px)

DESCRIPTION

Image indexing using edge-detection and Hausdorff distance computation.

Citation preview

  • 1. Lov, Ashish, Shikha, PrateekCSE09Guru Gobind Singh Indraprastha University

2. The application simulates a Visual search byaccepting an image from the user and indexinga specified directory of images according totheir similarity to the query image in terms ofshape, texture, color, etc. 3. Image similarity can be analyzed bycomputing a signature for each image to becompared A similarity index can be obtained bymeasuring the degree of approximationbetween the two given signatures The app defines independent signatures foreach feature of the image (color, texture) forcomparison and then combines the individualresults in order to obtain a better match 4. Aims at identifying points in animage where the image intensitychanges sharply or hasdiscontinuities This significantly reduces theamount of data and filters outuseless information whilepreserving the importantstructural properties in an image 5. Detects edges by looking for the maximumand minimum in the first derivative of theimage in gradient The actual algorithm involves thecomputation of the grayscale of the image (ifrequired) followed by the application of thegradient masks 6. Graph shows change in pixel Intensity aswe move from left to right in an image The edge shows the jump in intensity 7. Graph shows gradient of image Maxima represents local intensity change(Sobel method)Gradient 8. EXAMPLE: 2 9. Measures the degree of mismatch betweentwo finite point sets. Hausdorff distance is the maximum distanceof a set to the nearest point in the other set Given a set A of points a & a set B of pointsb; the Hausdorff distance from A to B is amaxi-min function given as: 10. The Hausdorff distance serves to check thedegree of similarity between images Lower the distance value, better the match This method gives good results, even inpresence of noise or occlusion Accounts for some basic geometrictransformations like: Translation Rotation Scaling 11. 1. Source Database2. Filtered images3. Hausdorff distances computed 12. Color Maps divide an image into blocks Blocks represent average pixel intensity of thecorresponding area of the image Generated by taking the Red, Green & Blueaverages of a block of pixels (16x16 in our app) Images scaled before map generation toimprove efficiency Images are compared by computing theabsolute difference of average intensitiesbetween respective blocks 13. A hash is a unique value of a fixed sizerepresenting a large amount of data (in thiscase, image data) Hashes of two images should match if andonly if the corresponding images also match Small changes to the image result in largeunpredictable changes in the hash 14. Secure Hash Algorithm (SHA) consists of aset of cryptographic hash functions ComputeHash method of SHA256Managedclass takes a byte array as an input parameterand produces a 256 bit hash of that data ImageConvertor class is used to convertImage (or Bitmap) objects from one data typeto another, such as a byte array By computing and then comparing the hashof each image, we can find exact matches 15. 1. Daniel P. Huttenlocher, Gregory A. Klanderman, and William J.Rucklidge. Comparing Images Using the Hausdorff Distance.IEEE Trans. Pattern Analysis and Machine Intelligence, 19932. J. Canny. A Computational Approach To Edge Detection.IEEE Trans. Pattern Analysis and Machine Intelligence, 19863. I. Sobel and G. Feldman. A 3x3 Isotropic Gradient Operator forImage Processing.Pattern Classification and Scene Analysis, 19734. H. Alt, B. Behrends and J. Blomer. Measuring the resemblance ofPolygon Shapes.Proc. Seventh ACM Symposium on Computational Geometry,19915. Herbert Schildt. C# 2.0: The Complete Reference, 2nd Edition.Tata McGraw-Hill, 20066. MSDN Library. msdn.microsoft.com/en-us/library/default.aspx