Sparse Representation-based Object Recognition - Sparse Representations and Compressive Sensing for Imaging and Vision - page 67

Fig. 5.3 Block diagram of
the method proposed in [112]
for the selection and
recognition of iris images

Input Iris Image

Iris Segmentation

Feature Extraction

Sparse Representation

Compute SCI

No

SCI > Threshold Reject Image

Yes

Compute
Reconstruction Error

Select
Minimizer

images suffer from high amounts of distortion. The recognition performance of the
SR based method for iris biometric [112] is summarized in Table 5.3 . As it can be
seen from the table SRC provides the best recognition performance over that of NN
and Libor Masek's iris identification source code [89].

5.3 Non-linear Kernel Sparse Representation

Linear representations are almost always inadequate for representing nonlinear data
arising in many practical applications. For example, many types of descriptors in
computer vision have intrinsic nonlinear similarity measure functions. The most
popular ones include the spatial pyramid descriptor [78] which uses a pyramid