12-04-2014, 02:11 PM
Co-occurrence matrix and its statistical features as a new approach for face recognition
Co-occurrence matrix .pdf (Size: 315.95 KB / Downloads: 52)
Abstract
In this paper, a new face recognition technique is introduced based on the gray-level co-occurrence matrix
(GLCM). GLCM represents the distributions of the intensities and the information about relative positions
of neighboring pixels of an image. We proposed two methods to extract feature vectors using GLCM for face
classification. The first method extracts the well-known Haralick features from the GLCM, and the second
method directly uses GLCM by converting the matrix into a vector that can be used in the classification
process. The results demonstrate that the second method, which uses GLCM directly, is superior to the
first method that uses the feature vector containing the statistical Haralick features in both nearest neighbor
and neural networks classifiers. The proposed GLCM based face recognition system not only outperforms
well-known techniques such as principal component analysis and linear discriminant analysis, but also has
comparable performance with local binary patterns and Gabor wavelets.
Introduction
Face recognition has been attracting the attention of the researchers as one of the most important techniques for
human identification. One of the limitations of real-time recognition systems is the computational complexity
of existing approaches. In the last couple decades, many systems and algorithms with high recognition rates
had been introduced. The general problem of these systems is their computational cost in data pre-preparation
and transformation to other spaces such as eigenspace [1, 2], fisherspace [3, 4], wavelet transform [5, 6] and
cosine transform [7].
Many researchers have used the gray-level co-occurrence matrix [8] for the extraction of features to be
used in texture classification. Gelzinis et al. [9] presented a new approach to exploiting information available in
the co-occurrence matrices computed for different distance parameter values.
Gray-level co-occurrence matrix
One of the simplest approaches for describing texture is to use statistical moments of the intensity histogram of
an image or region [16]. Using only histograms in calculation will result in measures of texture that carry only
information about distribution of intensities, but not about the relative position of pixels with respect to each
other in that texture. Using a statistical approach such as co-occurrence matrix will help to provide valuable
information about the relative position of the neighbouring pixels in an image.
Haralick features
In 1973, Haralick [8] introduced 14 statistical features. These features are generated by calculating the features
for each one of the co-occurrence matrices obtained by using the directions 0 ◦ , 45 ◦ , 90 ◦ , and 135 ◦ , then
averaging these four values. The symbol Δ , representing the distance parameter, can be selected as one or
higher. In general, Δ value is set to 1 as the distance parameter. A vector of these 14 statistical features is used
for characterizing the co-occurrence matrix contents. These features can be calculated by using the following
equations.
Simulation results and discussions
The simulations have been conducted on the four data sets: FERET, FRAV2D, Yale B and ORL face databases.
All experiments (except for FERET) were run randomly 10 times, after which results were averaged. The
standard test bed adopted in similar studies for the FERET database [17] has been used to test our algorithm
for identification. The results are reported in terms of the standard recognition rate. FERET database scenario
includes 200 people with 3 upright frontal face images each. In order to test the algorithms, two images of each
subject are randomly chosen for training, while the remaining one is used for testing.
ORL face database [18] is the second database we experimented with. It consists of 40 people with 10
images acquired for each one with different facial expressions and illumination variations.
FRAV2D Database [19] is the third database that is used, which comprises 109 people, each with 32
images. It includes frontal images with different head orientations, facial expressions and occlusions. In our
experiments, we used a subset of the database consisting 60 people with 32 images each with no overlap between
the training and test face images.
Conclusions
In this paper, we propose a new method, direct GLCM, which performs face recognition by using Gray-Level
Co-occurrence Matrix. The direct GLCM method is very competitive with state of the art face recognition
techniques such as PCA, LDA, Gabor wavelets, and LBP. Using smaller number of gray levels (bins) shrinks
the size of GLCM which reduces the computational cost of the algorithm and at the same time preserves the
high recognition rates. This can be due to the process of quantization which helps in suppressing the noise of
the images at higher grey levels. It is obvious from the results that the GLCM is a robust method for face
recognition with competitive performance.