International Journal of Science, Engineering and Technology Research (IJSETR) Volume 3, Issue 3, March 2014 Gender Recognition from Face Images with Weber Local Descriptor D.G.Agrawal , Pranoti M. Jangale Abstract: Gender Recognition by Face is an application of computer vision techniques to the problem of gender recognition that means the problem regarding genders of people presented in images or videos. This problem is solved by a 2-step process. The first step is to detect and localise human faces. This is achieved by a face detection algorithm. The second step is then to determine genders of those detected faces i.e. to separate his faces or her faces and is achieved by a gender classification algorithm Key words: dynamic textures, WLD descriptor, neural network, minimum distance measurement orientation. we represent an input image (or image region) with a histogram by combining the WLD feature per pixel. We call a WLD histogram hereinafter. Hence We call WLD;a dense descriptor. The proposed WLD descriptor employs the advantages of SIFT using the gradient and its orientation in computing the histogram, smaller support regions. and those of LBP in computational efficiency. But WLD differs from Local Binary Pattern and SIFT.There are 2 main steps involved in recognizing genders of humans presented in an image. These are face detection and gender classification, which are applied consecutively. I. Int ro duct io n II. Material and Methodology Over the past decades, there have been significant advances in facial image processing, especially, in a face detection area where a number of fast and robust algorithms have been proposed for practical applications. As a result, a number of research areas attempting to extend the works have been emerging, face recognition, facial expression recognition and gender recognition, for example. Since gender recognition can be considered as an extended work to face detection, this is why most research on gender recognition has focused on gender classification aspect and assumed the existence of face detection tools. With regard to gender classification, the techniques, tools and algorithms employed originate from fields such as computer vision, pattern recognition, statistics and machine learning. Weber Local Descriptor is a psychological law. It states that the change of a stimulus (such as lighting, sound) that we just notice is a constant ratio of the original stimulus. A human being would recognize it as background noise rather than a valid signal, When the change is smaller than this constant ratio of the original stimulus. The differential excitation component of the proposed Weber Local Descriptor (WLD) is computed for a given pixel. It is the ratio between the two terms: first is the intensity of the current pixel; the second is the relative intensity differences of a current pixel against its neighbours (e.g., 3 X3) square regions. We attempt to extract the local salient patterns in the input image, with the differential excitation component. In addition to this, current pixel’s gradient orientation is also computed. For each pixel of the input image, we compute two components of the WLD feature which are differential excitation and gradient 2.1 Face Detection The task achieved by face detection systems is to be understand using following steps. To know how to exploit uniqueness of faces in name recognition, the first step is to detect and localize those faces in the images. One of popular research areas is face detection in which many algorithms have been proposed for it. Considering the face detection as a binary classification task, most of them are based on the same idea. The task is to decide whether it is a face or not, given a part of image This is achieved by first transforming the given region into features and then using classifier trained on example images to decide if these features represent a human face. As faces show themselves having various sizes, appear in various locations and we also employ window-sliding technique. The idea in which the classifier classifies the portions of an image, at scales and all location, whether it is face or non-face. 2.2Gender Classification After faces are detected by face detection algorithm, they need to be decided if they are his or her faces. This is the task achieved by gender classification systems. Similar to the face detection task, the gender classification task is also considered as a binary classification problem but now with the result being male or female instead of face or non-face. 1 All Rights Reserved © 2012 IJSETR International Journal of Science, Engineering and Technology Research (IJSETR) Volume 1, Issue 1, July 2012 Essentially, gender classification consists of 4 main steps: pre-processing, feature detection, feature selection and classification. 2.2.1Pre-Processing Since, in real-life, it is unlikely that people will face directly and frontally towards the camera, face images often consist of some in-plane and out-of-plane rotations. Moreover, it is also unlikely that the light condition will be the same for all images. These variations greatly affect an accuracy of gender classifiers. The purpose of pre-processing step is thus to remove these variations as much as possible. Since not all the detected features are useful, the feature selection (or dimensionality reduction) module is employed here to choose only a subset of representative features. Doing feature selection not only gives us the relevant features and thus the more accurate result but also give us an additional advantage of faster computation time as the dimensionality of data is reduced. The popular feature selection techniques often employed in gender classification task are Prinicipal Component Analysis (PCA), Independent Component Analysis(ICA), Adaboost and Genetic Algorithm. 2.2.4Classification As with other computer vision applications, there is no unique solution to this problem. The common techniques involved in pre-processing step are face alignment, and light normalisation. Face alignment tries to align faces such that they are closed to a common or specified pose of face as much as possible, whereas light normalisation tries to get rid of the variation in illumination. One of the common employed normalisation techniques in the gender classification field is histogram equalisation. 2.2.2Feature Detection Working directly on raw pixel values can be very slow as one small face image can contain a thousand of pixels. Furthermore, not all the pixels will be useful. There can be an underlying structure that describes the differences between male and female faces better. Thus the feature detection module is employed here. Generally there are two types of features presented in the gender classification context, geometric-based features and appearance-based features. With all necessary features have been extracted, the final task is to decide whether or not those features represent female or male face. As there are obviously two decisions to make this is essentially binary classification task, that is, the classifier is trained on the female and male example face images so that it learns the decision boundary between these two classes. After that it uses what it learn to make a decision on the given face images. Among the binary classifiers, the most popular classifiers which give better performance than the others are a variation of Support Vector Machine (SVM), a variant of Adaboost and different Neural Network architectures. And among these classifiers, a number of comparative studies have been carried out and have suggested the best performance is obtained from the SVM. III. Results and Tables 3.1Results and analysis: Geometric-based features (also called local features) came from psychophysical explorations. They represent high-level face descriptions such as distances between nose, eyes and mouth, face width, face length, eyebrow thickness and so on Appearance-based features (also called global features) use low-level information about face image areas based on pixel values.Among appearance-based features the popular ones are: various texture features: e.g. Local Binary Pattern (LBP), Local Directional Pattern (LDP) and Pixel-Pattern-Based Texture Feature (PPBTF). histogram of gradients: Transform (SIFT) e.g. Scale-Invariant Fig: 1 Data Base Creation Feature coefficients of wavelet transformation of image: e.g. Gabor wavelet and Haar wavelet 2.2.3Feature Selection 2 All Rights Reserved © 2012 IJSETR International Journal of Science, Engineering and Technology Research (IJSETR) Volume 3, Issue 3, March 2014 Fig:5 Gender Classified Fig: 2 Browse Image 3.2Probabilistic Neural Network A PNN is predominantly a classifier since it can map any input pattern to a number of classifications. PNN is a fast training process and an inherently parallel structure that is guaranteed to converge to an optimal classifier as the size of the representative training set increases and training samples can be added or removed without extensive retraining. A consequence of a large network structure is that the classifier tends to be oversensitive to the training data and is likely to exhibit poor generalization capacities to the unseen data. In this paper, Probabilistic Neural Network is used to compare the features of input image with data base image which are obtained from the Local Binary Pattern. Fig: 3 Recognition Person Original/Fake Identification Fig: 4 Genders Classified for Template Data Base Conclusion In this paper we have implemented two techniques, one is to verify the signature whether authorized or unauthorized by measuring the Euclidean distance of both input image and data base images and we compared this results using principle component analysis (PCA). The second one is name identification using local binary pattern (LBP) and probabilistic neural network (PNN). Defining the effective features which results in minimum deviation for an signature instance may aid to further improvement of the system accuracy. An extension to the approach would be implementation of more accurate distance measurement techniques like minimum distance to verify the signature sample instead of Euclidean distance measure. Reference i.H. Cheng, Z. Liu, N. Zheng, and J. Yang, “A Deformable Local Image Descriptor,” Proc. IEEE Int’l Conf. Computer Vision and Pattern Recognition, 2008 ii. W. Zhao, R. Chellappa, P. J. Phillips and A. Rosenfeld, “Face recognition: a literature survey,” ACM Computing Surveys, vol. 35, pp. 399–458, December 2003. iii. M. Heikkila¨, M. Pietika¨inen, and C. Schmid, “Description of Interest Regions with Local Binary Patterns,” Pattern Recognition, vol. 42, no. 3, pp. 425-436, 2009 iv. M. S. Bartlett, J. R. Movellan, and T. J. Sejnowski, “Face recognition by independent component analysis,” IEEE Trans. Neural Networks, vol. 13, pp. 1450–1464, November 2002. v. J. F. Vargas, M. A. Ferrer, C. M. Travieso, J. B. Alonso: "Off-line signature verification based on grey level information using texture features," Pattern Recognition, vol. 44, no.2, pp. 375-385, 2011. vi. D. Bertolini, L.S. Oliveira, E. Justino, R. Sabourin, "Reducing forgeries in writer-independent off-line signature verification through ensemble of classifiers", Pattern Recognition, Vol. 43 pp.387-396, 2010. vii. A. Gilperez, F. Alonso-Fernandez, S. Pecharroman, J. Fierrez, J. Ortega- Garcia, "Off-line signature verification using contour features", Proceedings of the International Conference on Frontiers in Handwriting Recognition, ICFHR, 2008. viii .I. Siddiqi, N. Vincent, "Combining Contour Based Orientation and Curvature Features for Writer Recognition", 3 All Rights Reserved © 2012 IJSETR International Journal of Science, Engineering and Technology Research (IJSETR) Volume 1, Issue 1, July 2012 Lecture Notes in Computer Science, Volume 5702, pp. 245-252, 2009 ix.M. Szummer and R.W. Picard, “Temporal Texture Modeling,” Proc. IEEE Conf. Image Processing, vol. 3, pp. 823-826, 1996 x. J. Chen, G. Zhao, and M. Pietikäinen, “An improved local descriptor and threshold learning for unsupervised dynamic texture segmentation,” in Proc. 12th IEEE Int. Conf. Comput. Vis. Workshop, Oct. 2009, pp. 460–467. . . 4 All Rights Reserved © 2012 IJSETR