Direct kernel biased discriminant analysis: a new content-based image retrieval relevance feedback algorithm
read more
Citations
Statistical Pattern Recognition
A survey of graph edit distance
A comprehensive review of current local features for computer vision
Gabor-Based Region Covariance Matrices for Face Recognition
Marginal Fisher Analysis and Its Variants for Human Gait Recognition and Content- Based Image Retrieval
References
The Nature of Statistical Learning Theory
Matrix Analysis
Content-based image retrieval at the end of the early years
Color indexing
Texture features for browsing and retrieval of image data
Related Papers (5)
Nonlinear dimensionality reduction by locally linear embedding.
Frequently Asked Questions (10)
Q2. What have the authors stated for future works in "Direct kernel biased discriminant analysis: a new content-based image retrieval relevance feedback algorithm" ?
In the future, the authors plan to generalize the tuning method to select the parameters of kernel-based algorithms. For RF in CBIR, the training size of the training set is small, so the leave-one-out method to tune the parameters can be used.
Q3. Why is DKBDA motivated by the fact that face recognition has made some advances?
DKBDA is motivated by a) the fact that direct LDA (DLDA) [23], [26], recently developed for face recognition, has made some advances; and b) unlike face recognition, image retrieval deals with diverse images, so the nonlinear properties of imagefeatures should be considered because of the success of kernel algorithms in pattern recognition.
Q4. How many iterations are required for the SVM RF?
In the case of the top ten results, after four iterations, the precision of the proposed DBDA is already higher than 90% while seven iterations are required for the BDA algorithm and more than nine iterations for the SVM RF.
Q5. What are the main features of a CBIR system?
Generally in a CBIR RF system images are represented by the three main features: color [3], [4], and [10]–[12], texture [5]–[10], [12], and shape [11]–[13].
Q6. What is the feedback process for the image retrieval system?
the user provides feedback by clicking on the “thumb up” or “thumb down” button according to his/her judgment of the relevance of the sorted images.
Q7. What is the way to solve the SSS problem?
Zhou et al. [24], [25] solve the SSS problem by the regularized version and , which adds small quantities to the diagonal of the scatter matrices.
Q8. What is the SSS problem for LDA?
LDA has the SSS problem when the number of the training samples is smaller than the dimension of the low-level visual features, which is almost always true for CBIR RF.
Q9. How does the computer do the relevance feedback?
In their experiments, the computer does the relevance feedback iterations automatically without mislabeled samples using the 80 concept groups described previously.
Q10. How is the DBDA performance shown in Fig. 4?
IDKBDA is proved to be of approximately the same capabilities as DKBDA, but it can speed up the DKBDA remarkably by saving about 20% of the running time (9 and 11 h for all the 300 queries and nine iterations for each query).