Age Group Determination from Face Using Texture Classification based on Probabilistic Non-Extensive Entropy

Abstract

Various types of changes occur in face as the age of a person progresses from childhood to senile stage. One of most prominent change that happens is the gradual change in the texture of the face. Such changes are highly conspicuous in the form of wrinkles on the face. As a person grows old, the texture of the skin changes and is initially most visible in the face. It may start with introduction of fine lines and dark spots progressing all the way to high intensity of wrinkles on the face. It is a known fact that texture is an integral part of natural images and has been widely used for solving may problems in the area of pattern recognition, image processing and computer vision. Texture classi?cation aims at assigning labels to the unknown textures represented by some textural features. The technique presented here is suitable for classification of people into different age groups by texture characterisation of the face. A probabilistic non-extensive entropy feature based on a Gaussian information measure has been used for texture characterization. This entropy is bounded by ?nite limits and is non-additive in nature. It can be used to represent the information content in non-extensive systems having some degree of regularity or correlation thereby finding its application in texture classi?cation problems since textures found in nature are random and at the same time contain some degree of correlation or regularity at some scale. The probabilistic non-extensive entropy used for age group determination based on texture classification is primarily founded upon Gaussian information gain function. The non-additive property of this entropy makes it especially significant for the texture classi?cation required for age determination. Also the non-linearity of the information gain function plays an important role in the identi?cation of textures having high spatial correlation and containing non-additive information content. This entropy measure increases the non-linearity of the exponential information gain introduced in by replacing the linear exponent of the exponential by a quadratic probability term. Inner Product Classifier is used for performing classification. It considers the errors between the training features and the test image features using triangular or t-norms. The triangular norms aid in highlighting the errors and find a margin between them. The inner product between the aggregated training features vector and t-norm of the error vectors must be the lowest for the test feature vectors for it to match with the training feature vectors.