Local Derivative Pattern Versus Local Binary Pattern: Face Recognition With High-Order Local Pattern Descriptor
read more
Citations
Local Tetra Patterns: A New Feature Descriptor for Content-Based Image Retrieval
Local Directional Number Pattern for Face Analysis: Face and Expression Recognition
Learning Compact Binary Face Descriptor for Face Recognition
Discriminative Multimanifold Analysis for Face Recognition from a Single Training Sample per Person
Local binary features for texture classification
References
Eigenfaces for recognition
Multiresolution gray-scale and rotation invariant texture classification with local binary patterns
Eigenfaces vs. Fisherfaces: recognition using class specific linear projection
A comparative study of texture measures with classification based on featured distributions
Face recognition: A literature survey
Related Papers (5)
Enhanced Local Texture Feature Sets for Face Recognition Under Difficult Lighting Conditions
Frequently Asked Questions (15)
Q2. What have the authors stated for future works in "Local derivative pattern versus local binary pattern: face recognition with high-order local pattern descriptor author" ?
Investigation and comparison on mutliscale versions of LDP and LBP are interesting future work to fine tune the proposed approach.
Q3. How long did it take to match the LBP and LDP?
The computational time for one-to-one image matching (including feature extraction and feature matching) of LBP and the third-order LDP were 0.054 and 0.180 s, respectively.
Q4. What are the main contributions of this paper?
The main contributions of this paper include: 1) A novel local descriptor, high-order Local Derivative Pattern, is proposed as an object descriptor.
Q5. Why is the proposed high-order local pattern descriptor applicable to other object recognition tasks?
Due to its excellent performance, the authors expect that the proposed high-order local pattern descriptor is applicable to other object recognition tasks as well.
Q6. How did the authors normalize and crop the frontal face images?
Considering that images in FRGC are of higher resolution, the authors normalized and cropped them to 128 168 pixels to evaluate the performances in a higher resolution than in previous experiments.
Q7. How were the eye locations of the FERET probe images shifted?
In this experiment, the two eye locations of the FERET probe images were shifted by four independent random values, i.e., displacements of left and right eyes in and directions, which were gener-ated using random Gaussian distribution with ranging from 1 to 3 by rounding them off to the nearest integers.
Q8. What is the effect of the noise on the detection accuracy of LBP and LDP?
When increases further, the identification accuracy curves of both LBP and LDP become flatter, showing that the large subregions are more robust to large misalignments than smaller ones.
Q9. How does the LBP operator label the pixels of an image?
The original LBP operator labels the pixels of an image by thresholding the 3 3 neighborhood of each pixel with the value of the central pixel and concatenating the results binomially to form a number.
Q10. What is the simplest way to extend the LDP to feature images?
it is anticipated that extending the proposed high-order local pattern description to feature images containing wider range of appropriate discriminative features could achieve a higher level of system performance.
Q11. Why does the fourth-order perform better than the gray-level images?
Different from the results on the gray-level images, the performance of the fourth-order degrades gracefully, probably because Gabor features contain high-order discriminative information and less sensitive to noise due to kernel convolving operation [6], [15].
Q12. What is the average recognition rate of LBP and LDP?
The average recognition rates on the four probe sets against different of Gaussian noise are illustrated in Fig. 9, showing that LDP maintains a 13.7% to 15.0% higher accuracy over LBP when increases up to 3, and then its accuracy drops greater than that of LBP with larger amount of noise.
Q13. How do you calculate the second-order directional local derivative pattern?
To calculate the second-order directional Local Derivative Pattern, , in direction at , the four templates in Fig. 3(a) are applied on the image by aligning and to , respectively.
Q14. What is the difference between the LBP method and the other methods?
Different from statistic learning methods tuning a large number of parameters, the LBP method is very efficient due to its easy-to-compute feature extraction operation and simple matching strategy.
Q15. How do the authors calculate the third-order Local Derivative Pattern?
To calculate the third-order Local Derivative Pattern, the authors first compute the second-order derivatives along 0 , 45 , 90 and 135 directions, denoted as where , 45 , 90 , 135 .