Download presentation
Presentation is loading. Please wait.
1
Classification techniques
OLAM SE Petr Chmelař Lukáš Stryka Koncepty pro sledování chodců
2
Abstract My research aims to the knowledge discovery in multimedia data, especially mining visual data in large databases and its semantic description. My proposed contribution to the CPG project at FIT BUT is the application of corners and invariant region detectors, detection of surfaces and objects in addition to the proposed techniques, the comparison of classification techniques for hardware acceleration of recognition, tracking and augmented reality. Abstrakt anglicky
3
Projection of 3D objects’ surfaces into 2D
Texture Analysis Projection of 3D objects’ surfaces into 2D Obrázek 3.2: Příklady textur z [Br66] – alba P. Brodatze
4
Texture analysis Texture synthesis
Texture description (feature extraction) Classification (into known classes) Image segmentation (unknown number of classes) region-based boundary based Shape / object detection Many methods Statistical (Haralick) Signal processing (Gabor Filters) Geometry (Markov models) Efficient in image content indexing for content-based retrieval in large DBs.
5
Gabor Filters
6
DM applications Frequent set (association rules) application
Irregularities, … noise: How to write such an algorithm?
7
Classification Theory of Information state class observation parameter
estimation
8
Bayessian Classification
Bayesian provides a method for adjusting degrees of belief of new information. likelihood prior information marginal probability P(y | xi)P(xi) posterior probability
9
Non-naïve Bayessian Classification
Association analysis as a non-naïve Bayessian classification. a1 a2 … ap bp+1 … bq (+ Support, Confidence, Correlation, Cover) maximum likelihood estimation maximum a posteriori
10
Find linear separation (hyper-plane):
SVM Find linear separation (hyper-plane):
11
Kernels
12
Comparison MIT-CBCL face dataset (2 x 2500, Haar features)
13
Questions… ? Thank you. ctrl + S
14
References HAN, J., KAMBER, M. Data Mining: Concepts and Techniques ISBN CRISITIANI, N. Kernel Methods for Pattern Analysis Please mail for more literature & self-citations.
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.