Face Recognition By Elastic Bunch Graph Matching Pdf

• • Part of the book series (LNCS, volume 1296) Abstract We present a system for recognizing human faces from single images out of a large database with one image per person. The task is difficult because of image variation in terms of position, size, expression, and pose. The system collapses most of this variance by extracting concise face descriptions in the form of image graphs. In these, fiducial points on the face (eyes, mouth etc.) are described by sets of wavelet components ( jets). Image graph extraction is based on a novel approach, the bunch graph, which is constructed from a small set of sample image graphs.

Face Recognition by Elastic Bunch Graph Matching* Laurenz Wiskott it, Jean-Marc Fellous 2~, Norbert Kriiger 1, and Christoph vonder Malsburg 1'2. FACE RECOGNITION USING ELASTIC BUNCH GRAPH MATCHING Sandeep R1, D Jayakumar2 Dept. A face recognition system is a computer application for.

Recognition is based on a straight-forward comparison of image graphs. We report recognition experiments on the FERET database and the Bochum database, including recognition across pose.

Contents • • • • • • • • • • • • • • • • • • • • • • Introduction Elastic Graph Matching (EGM) is a biologically inspired algorithm for object recognition in the field of computer vision. It draws its biological inspiration from two sources: (i) The visual features used are based on, which have been found to be a good model of early visual processing in the, more precisely in primary. (ii) The matching algorithm itself is an algorithmic version of dynamic link matching (DLM), which is a model of invariant object recognition in the brain. Visual objects in EGM are represented as, where the nodes represent local textures based on Gabor wavelets and the edges represent distances between the node locations on an image. Thus an image of an object is represented as a collection of local textures in a certain spatial arrangement.

If a new object in an image shall be recognized, the labeled graphs of stored objects, so-called model graphs, are matched onto the image. For each model graph the locations for the nodes in the image are optimized such that local texture of the image fits the local texture of the model and distances between the locations fit the distances between the nodes of the model. The model graph with the best fit constitutes the recognized object, and with its node locations in the image an image graph can be created. Elastic Bunch Graph Matching (EBGM) is an extension to elastic graph matching for object classes with a common structure, such as faces in identical pose.

All instances of such a class are represented by the same type of graph. From these graphs a bunch graph of same structure is created, with the nodes representing local textures of any object in the class, e.g. All variants of a left eye, and the edges represent average distances between the node locations, e.g. The average distance between the two eyes. This permits taking advantage of the combinatorics of the local textures to represent instances of the object class not seen before.

For instance, the textures of the eyes could be taken from one face and the textures of the mouth from another face to represent a new face that shares features with the two stored faces. Thus, a bunch graph is an abstraction for representing object classes rather than individual objects.

EBGM can only be applied to objects with a common structure, such as faces in frontal pose, sharing a common set of landmarks like the tip of the nose or the corner of an eye. For the recognition of arbitrary objects, in the absence of landmarks, the graphs are required to be dynamic with respect to both shape and attributed features. To this end, have proposed a graph that lets generic object representations, model or bunch graphs, emerge from a collection of arbitrary objects. The idea is to extract typical local texture arrangements from the objects and provide the rules to compose them as needed to represent new objects.

Algorithm Gabor Wavelets. Figure 6: Creation of Jets by a Gabor wavelet transform.

Upper left: The circles indicate the Gaussians in the frequency domain that represent the complex Gabor wavelets. Bottom: At a given image location each wavelet yields a complex response, only the amplitude of which is illustrated here by the gray values. Upper right: The illustrated amplitude responses can be stacked into a more compact shape to represent a jet. The Gabor wavelet transform yields a value for each wavelet at all locations of the image. Symantec Backup Exec 2014 Crack For Gta.

Thus, with the standard parameters and discretized images it yields 80 (40 real + 40 imaginary) values at any pixel position. This set of values for a single pixel position is referred to as a jet (J ) (Figure ). Since a jet contains values from wavelets of different frequency and orientation, one can think of it as a local Fourier transform, and it is as such a representation of the local texture. It is in fact possible to reconstruct the image gray values from a jet in a small surrounding of its location, except for the mean value.