C. Liu and H. Wechsler, "Robust coding schemes for indexing and retrieval from large face databases," IEEE Trans. on Image Processing, vol. 9, no. 1, pp. 132--137, 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....this increase in digital content is a need for database management tools that will allow people to easily archive and retrieve desired content from their digital collections. Furthermore, since humans and their activities are typically the subjects of interest in both images and videos [2], 3] 5] detection and identification of human faces will help to automate image and video archival based on semantic (high level) concepts, such as the face and facial components. This will allow us to search a database using queries of the form find all the images containing John s faces, ....

C. Liu and H. Wechsler, "Robust coding schemes for indexing and retrieval from large face database," IEEE Trans. Image Processing, vol. 9, no. 1, pp. 132--137, 2000.

....However, the robustness of the FLD procedure depends on whether or not the within class scatter can capture enough variations for a specific class. When the training sample size for each class is small, the FLD procedure leads to overfitting, and hence with poor generalization to new data [12] [13]. Another recently proposed method for face feature extraction is Independent Component Analysis (ICA) 1] which separates the high order moments of the input in addition to the second order moments [3] However, it is not clear that how much is the non Gaussianity of the face images and how ....

....to select the features. In standard eigenface and the PRM approaches, the features derived from PCA are sorted in descending order according to the eigenvalues of the principal components. The higher the dimensions, the smaller the eigenvalues, as is the case of traditional approaches [12] [13]. It is obvious that under the pairwise recognition framework, the AdaBoost (labeled as PairBoost) and Bayes (labeled as PairProb) approaches have much higher recognition accuracies than the standard eigenfaces and PRM in the low dimensions (# # ##) In Fig. 2, the feature dimensions start from #, ....

[Article contains additional citation context not shown here]

C. Liu and H. Wechsler. Robust coding scheme for indexing and retrieval from large face database. IEEE Trans. Image Processing, 9(1):132--137, 2000.

No context found.

C. Liu and H. Wechsler, "Robust coding schemes for indexing and retrieval from large face databases," IEEE Trans. on Image Processing, vol. 9, no. 1, pp. 132--137, 2000.

No context found.

C. Liu and H. Wechsler, "Robust coding schemes for indexing and retrieval from large face databases," IEEE Trans. on Image Processing, vol. 9, no. 1, pp. 132--137, 2000.

No context found.

C. Liu and H. Wechsler, "Robust coding schemes for indexing and retrieval from large face databases," IEEE Trans. on Image Processing, vol. 9, no. 1, pp. 132--137, 2000.

No context found.

C. Liu and H. Wechsler, "Robust coding schemes for indexing and retrieval from large face databases," IEEE Trans. on Image Processing, vol. 9, no. 1, pp. 132--137, 2000.

No context found.

C. Liu and H. Wechsler, "Robust coding schemes for indexing and retrieval from large face databases," IEEE Trans. on Image Processing, vol. 9, no. 1, pp. 132--137, 2000.

No context found.

C. Liu and H. Wechsler, "Robust coding schemes for indexing and retrieval from large face databases," IEEE Trans. on Image Processing, vol. 9, no. 1, pp. 132--137, 2000.

....The corresponding but reduced shape and texture features are then combined through a normalization procedure to form the integrated shape and texture features. The dimensionality reduction procedure, constrained by the Enhanced FLD (Fisher Linear Discriminant) Model (EFM) for better generalization [17], maintains a proper balance between the spectral energy needs of PCA for adequate representation, and the Fisher Linear Discriminant (FLD) requirements that the eigenvalues of the within class covariance matrix should not include small trailing values as they tend to encode noise and appear in ....

.... the Most Discriminating Features (MDF) 21] and the Fisherfaces [1] The combined use of PCA and FLD like methods is an improvement over PCA methods, but still has its own drawbacks, especially those associated with overfitting and lack of generalization as a result of insufficient training data [17], 8] One can show that the MDF space is, however, superior to the PCA space for face recognition, only 4 when the training images are representative of the range of face (class) variations; otherwise, the performance difference between the PCA and MDF spaces is not significant [21] To further ....

[Article contains additional citation context not shown here]

C. Liu and H. Wechsler, "Robust coding schemes for indexing and retrieval from large face databases," IEEE Trans. on Image Processing, vol. 9, no. 1, pp. 132--137, 2000.

....it does not distinguish the different roles of the within and the between class variations, and it treats them equally. This should lead to poor testing performance when the distributions of the face classes are not separated by the mean difference but instead by the covariance difference [15] [30], 27] High variance by itself does not necessarily lead to good discrimination ability unless the corresponding distribution is multimodal and the modes correspond to the classes to be discriminated. One should also be aware that as PCA encodes only for 2nd order statistics, it lacks phase and ....

C. Liu and H. Wechsler, "Robust coding schemes for indexing and retrieval from large face databases," IEEE Trans. on Image Processing, vol. 9, no. 1, pp. 132--137, 2000.

....full anticaricature [13] 3. Face Representation and Classification Robust indexing and retrieval encoding schemes require both low dimensional feature representations, for data compression and faithful reconstruction purposes, and enhanced discrimination abilities for subsequent image retrieval [31]. Such complementary requirements can be addressed using an approach similar to constrained optimization, where the unifying theme is that of lowering the subspace dimension ( data compression ) subject to increased fitness for the discrimination index. The three candidates for possible coding ....

.... face recognition, only when the training images are representative of the range of face (class) image variations; otherwise, the performance difference between the MEF and MDF is not significant [39] The FLD procedure, when implemented in a high dimensional PCA space, leads often to overfitting [31]. Overfitting is more likely to occur for the small training sample size scenario, which is the typical one for face recognition [35] One possible remedy for this drawback is to artificially generate additional data and thus increase the sample size [18] Another solution, to analyze the reasons ....

[Article contains additional citation context not shown here]

C. Liu and H. Wechsler. Robust coding schemes for indexing and retrieval from large face database. IEEE Trans. Image Processing, 9(1):132--137, 2000.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC