Global, semi-global and local color angular features for unsupervised face recognition

• Autor(a) principal: Chidambaram, Chidambaram
• Data de Publicação: 2003
• Outros Autores: Lopes, Heitor Silvério, Vieira Neto, Hugo
• Tipo de documento: Artigo de conferência
• Idioma: eng
• Título da fonte: Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT))
• Texto Completo: http://repositorio.utfpr.edu.br/jspui/handle/1/648
• topic: Percepção facial; Percepção de padrões; Visão por computador; Cor; Face perception; Pattern perception; Computer vision; Color

In face recognition applications, dealing with images under different conditions is a challenging task because they can affect dramatically the recognition performance. Among many image features, color is an useful feature which is generally used for image matching and retrieval purposes. Besides, to represent images through features, we generally need an extensive number of parameters forming a large feature set. Color angles need only three parameters to represent an image in a small feature set and are considered as pose and illuminant-invariant. Hence, in this work, we have made an attempt to study the use of color angles in face recognition approach with images obtained under different conditions. In addition to this, face image features are spatially extracted from different combination of sub-images similar to the edge histogram descriptor scheme denominated as Global, Semi-Global and Local features. Since we have proposed an unsupervised learning approach, no previous knowledge about images are required. Six types of images obtained under two different illumination conditions including with face expression and scale are used as query images in a base of images obtained under controlled condition. According to the experimental results, an expressive recognition rate can be obtained from face expression and scale. One of the main goal of this work is the use of Semi-Global features with Global and Local features. From this initial study, we can identify that the Local and Semi-Global features influence in the recognition performance than Global features.