A two level approach for scene recognition
read more
Citations
Machine learning
Landmark recognition with compact BoW histogram and ensemble ELM
Landmark recognition with sparse representation classification and extreme learning machine
Landmark recognition with sparse representationclassification and extreme learning machine
References
Maximum likelihood from incomplete data via the EM algorithm
Textural Features for Image Classification
Bagging predictors
Machine learning
Related Papers (5)
Frequently Asked Questions (13)
Q2. What contributions have the authors mentioned in the paper "A two level approach for scene recognition∗" ?
In this paper, the authors present a stratified approach to both binary ( outdoor-indoor ) and multiple category of scene classification. The authors then extract some very simple features from those PDRMs, and use them to train a bagged LDA classifier for 10 scene categories. To test this classification system, the authors created a labeled database of 1500 photos taken under very different environment and lighting conditions, using different cameras, and from 43 persons over 5 years.
Q3. How do the authors get a set of LDA scene classifiers over these feature vectors?
By employing the random subspace method [12, 28] and bootstrapping [31], the authors obtain a set of LDA scene classifiers over these feature vectors.
Q4. How do the authors create multiple LDA classifiers?
To improve the classification rate, the authors have implemented variations on random subspace generation [12, 28] and bootstrapping [31] to create multiple LDA classifiers.
Q5. What is the simplest way to compute the density of a material class?
For any given photo, the authors scan local image patches, extract their color-texture feature vector, normalize each of its components from 0 to 1 [1], project it to the lower dimensional subspace Z computed by LDA, and finally compute the density value given by equation (1) for all 20 material classes.
Q6. How do the authors evaluate the membership density of the image patches for each class?
Once the authors obtain 20 Gaussian mixture models {πik, P (z; θik), i = 1, 2, ..., 20} for 20 material classes, the authors can evaluate the membership density values of image patches for each material class.
Q7. How do the authors prepare the training data?
To prepare the training data, the authors manually crop image regions for each material in their database, and randomly draw dozens of 25 by 25 pixel patches from each rectangle.
Q8. How many mixtures are used for each class?
The number of mixtures gc and the model parameters {πck, θck} for each material class c are initialized by spectral clustering [21] and learned in an iterative Expectation-Maximization manner [31, 7] where gc ranged from 4 to 8 depending on the material class.
Q9. What is the classification method for indoor-outdoor scenes?
An misclassified outdoor photo.moment features of PRDMs are useful in outdoor scenes, but reduce the recognition rate for indoor scenes.
Q10. What is the way to evaluate the texture discrimination performance of Haralick?
The authors evaluate their texture discrimination performances experimen-2The reference and neighbor pixel intensities normally need to be quantized into 16 or less levels instead of 256 which results in not too sparse GLCM.tally in section 4 and find Haralick features generally perform better.
Q11. What is the common method used to find the marginal distributions of the basic filter banks?
For texture modeling, Zhu et al [35] pursue features to find the marginal distributions which are also the linear combinations of the basic filter banks, but they use a much more complex method (Monte Carlo Markov Chain) to stochastically search the space of linear coefficients.
Q12. What is the way to separate data from different classes?
The LDA computation is reviewed in appendix B.When each class has a Gaussian density with a common covariance matrix, LDA is the optimal transform to separate data from different classes.
Q13. What is the combination of LDA and Gaussian mixture models?
the authors describe a combination of LDA and Gaussian mixture models that achieves a good balance of discrimination and smoothness.