Face recognition using Collaborative Representation based Classification

The Common View-Based Classification (CRC) method for face recognition is more accurate and efficient. The common representation encodes the y signal on the complete dictionary Φ so that y≈Φα and α are sparse vectors SRC α is often specified by the size l۱, which leads to the SRC coding model (min) -αǁαǁ-۱ s.t.ǁy-Фαǁ- ۲<ε.Becomes That £is, a small constant the main idea of this method is to encode the test sample on a dictionary under SRC it is also classified based on the coding vector. X(i)Є R(m*n) A set of training examples from class i is specified. So that each column Xi is an example.It is assumed that there is a class K of members and X =[x۱,x۲,…,x۲] for an image of the desired face yЄRm, which is coded on X. ۱-Y ≈ Xα that in α=[ α۱ ;…, αi;…; αK] ] and αi، Related vector Xi is if y is of class i, the relation y≈Xiαi is usually well maintained This means that most of the coefficients in αk, k ≠ i are small And only αi has significant values in other words, thin non-zero inputs in α can encode y in SRC, it is assumed that the face images are aligned and methods have been proposed to resolve image alignment differences or the problem of changing their position. Minimizing the l۱ required in SRC-based pattern classification can be costly. Many methods try to improve the setting of l۱ on the α coding vector and Frequently use tutorial examples, taken from all classes to display examples y relatively ignored. In the method based on the common representation of size l۱ and l۲ in order to specify the coding vector α and the rest of the coding e=y−Xα is used. The characteristic e in size l۱ or l۲ is related to the robustness of the method relative to the throw pixels. While the characteristic α in the soft l۱ or l۲ is related to the distinction of the characteristic y in the face when the image of the face is not obstructed or damaged, the size l۲ is sufficient to model e the distinction of the y feature in the face is often related to its dimensions. If the dimensions and distinction y are large. The coding coefficients α will be naturally thin and will focus on samples with a class label similar to y. It does not matter if l۱ or l۲ is used to set α. When the dimensions y are too small. In this case, adjusting the size of l۱ to α empties α, which in turn improves its differentiation.