Face expression

Characteristics

  • The way of coding is better in comparison to the others states-of-the-arts.
  • It is more robust against noise, illumination, and age condition.

Full description

Face expression recognition is widely used for making research in this field, due to is a god challenge to can recognize the expression of the persons, and can know their humor, e. g. if we know the humor of the people we can know how to talk to them, or whatever we want to do.

The Local Derivative Pattern (LDP) approach assign eight bit binary code to each pixel of an input image. This pattern is then calculated by comparing the relative edge response values of a pixel in different directions. The Kirsch, Prewitt, and Sobel edge detectors are some of the different representative edge detectors which can be used in this regard. Using any mask describe before we can capable to detects different directional edge responses more accurately

than the others because it considers all 8 neighbors. Given a central pixel in the image, the eight-directional edge response values \({m_i}\), \(i\) = 0, 1,…, 7 are computed by Kirsch masks, \(M_i\), in eight different orientations centered on the pixel’s position.

The response values are not equally important in all directions. The presence of a corner or an edge shows high response values in some particular directions. Therefore, we need to know the most prominent k directions to generate the LDP. Here, the top-k directional bit responses, bi, are set to 1. The remaining 8 k bits of the 8 bit LDP pattern are set to 0. Finally, the LDP code is derived by

\[ \text{LDP}_k = \sum_{i=0}^7 b_i(m_i-m_k) \times 2^i, \quad b_i(a) = \begin {cases} 1, &a \geq 0, \\ 0, &a < 0, \end {cases} \]

where \(m_k\) is the k-th most significant directional responses, represent equation 1.

mat0

The image (a)  shows the eight directional edge responses position and (b) represent the LDP binary bit position

mat1

The figure shows an example of like come up with the 8 directional responses using the equation 1. In the Table shows the recognition rate of the proposed method, in two different data sets.

mat2

And Local Directional Pattern Variance, generally, texture can be well represented when characterized by a spatial structure along with its contrast. The LDP feature only contains the distribution of local structures. A low contrast structure contributes equally with a high contrast one in the LDP histogram. However, texture with significant contrast should impact more since human eyes are more sensitive to high contrast regions. Hence, we account for the contrast information within the feature descriptor. The variance of a structure is related to the texture. Generally, high frequency texture regions have higher variances and contribute more to the discrimination of texture images. Therefore, the variance σ is introduced as an adaptive weight to adjust the contribution of the LDP code in the histogram generation. The proposed LDPv descriptor is computed as

\[\begin{align} \text{LDP}v(\tau) &= \sum_{r=1}^M \sum_{c=1}^N \omega(LDP_k(r,c), \tau) \\ \omega(\text{LDP}_k(r,c), \tau) &= \begin {cases} \sigma(\text{LDP}_k(r,c)) &\text{LDP}_k(r,c) = \tau\\ 0 &otherwise \end {cases} \\ \sigma(\text{LDP}_k(r,c)) &= \frac{1}{8} \sum_{i=0}^7 (m_i-\overline{m})^2 \end{align}\]

where \(\overline{m}\) is the average of all directional responses \({m_i}\) calculated for position \((r,c)\). When LDP and variance \(\sigma\) are treated as the two orthogonal axes in a coordinate system, the LDPv can be regarded as the integral projection along the \(\sigma\) axis.

Different techniques have been proposed to classify facial expressions. A comparative analysis of four machine learning technique, namely Template matching, Linear Discriminant Analysis (LDA), Linear programming, and SVM, which showed SVM performed the best. However, template matching is commonly adopted for its simplicity. In this section, both template matching and SVM are explained for expression classification based on LDPv features.

In the following table shows the results, testing with two different data sets,  where we can see that the recognition rate is better than the other state-of-the-arts.

mat3

References

SCI/SCIE Indexed Journal:

▷ Taskeed Jabid, Md. Hasanul Kabir, and Oksam Chae, “Robust Facial Expression Recognition based on Local Directional Pattern”, ETRI Journal, Volume 32, No. 5, October, 2010. (SCI) .
▷ Md. Hasanul Kabir, Taskeed Jabid, and Oksam Chae, “Local Directional Pattern Variance (LDPv): A Robust Feature Descriptor for Facial Expression Recognition”, International Arab Journal of Information Technology (IAJIT), 2012. (SCIE).

Support

This work was supported by the Korea Research Foundation Grant funded by the Korean Government (KRF-2010-0015908).

Comments are closed.