3. Classification based on SRC algorithm
Given a new test sample y from one of the classes in the
training set, we first compute its sparse representation:
α^ 0 ¼ argminx j jαj j 0 s:t: y ¼ Dα ð13Þ
where α0 ¼ ½0; ⋯; 0; wi;1; ⋯; wi;ni
; 0; ⋯; 0
T AIRn is a coefficient vector
whose entries are zero except those associated with the i-th
class, j j∙j j 0 denotes the l0-norm, which counts the number of
nonzero entries in a vector.
The problem of finding the sparsest solution is NP-hard which can
be solved by using matching pursuit [30]. In this paper, we recover y
by solving the following reduced l1-minimization problem:
α^ 1 ¼ argminj jαj j 1 s:t: y ¼ Dα ð14Þ
For each class i, let δi : IRn-IRn be the characteristic function
that selects the coefficients associated with the i-th class. For
αAIRn, δiðαÞAIRn is a new vector whose only nonzero entries are
the entries in α that are associated with the i-th class, one can
approximate the given test sample y as y^ ¼ Dδiðα^Þ. Finally, we can
classify y based on these approximations by assigning it to the
object class that minimizes the residual between y and y^:
min rðyÞ¼jj yy^j j 2 ð15Þ
For comprehending the categories of a test sample, the algorithm’s
classification process is as shown in Fig. 3, meanwhile
Algorithm 2 summarizes the complete recognition procedure.
Algorithm 2. The algorithm of face recognition based on SRC
Face recognition based on discriminative dictionary learning
and sparse representation
1. Convolve the face image function fðzÞ with the filter function
Ψμ;vðzÞ to get the Gabor images;
2. Calculate the ULBP histograms of the 40 Gabor images;
3. Gain the better dictionary for sparse representation by dictionary
learning;
4. Normalized the training dictionary;
5. Solve the l1-minimization problem:x^ 1 ¼ argminx j j xj j 1
s:t: y ¼ Dx
6. Compute the residuals:riðyÞ¼jj yDδðx^1Þj j 2
F