Classi cation of Edges
نویسندگان
چکیده
Structural classiication of edges is investigated by interpretation of visual motion from uncalibrated stereo views. Local surface patches close to edges are modeled by a planar description to simplify the determination of their associated mappings between the two projected views. The projectivity between the corresponding patches in the stereo images is parameterized under epipolar constraints and optimized by correlation scoring. From the diierences in the resulting mappings of the two patches at the side of the edge a rule base is outlined under statistical estimation of the uncertainty in the classiier. Three types of edges are successfully discriminated: surface markings, face junctions, and occluding object boundaries.
منابع مشابه
Algorithm and VLSI Design of a Feature-Based Classi ed Vector Quantizer for Image Coding
In this paper, a feature-based classi ed vector quantization (FCVQ) algorithm and VLSI implementation of the classi er are presented. The FCVQ technique exploits the characteristics of discrete cosine transform (DCT) and the concepts of block truncation coding (BTC) to simplify the classi cation and preserve the edge information e ciently. During the classi cation, an input block is classi ed a...
متن کاملSegmentation and Classi cation of Edges Using Minimum Description Length Approximation and Complementary Junction Cues
This article presents a method for segmenting and classifying edges using minimum description length (MDL) approximation with automatically generated break points. A scheme is proposed where junction candidates are rst detected in a multi-scale preprocessing step, which generates junction candidates with associated regions of interest. These junction features are matched to edges based on spati...
متن کاملImproving classical contextual classi ® cations 2
This paper shows some combinations of classi® ers that achieve high accuracy classi® cations. Traditionally the maximum likelihood classi® cation is used as an initial classi® cation for a contextual classi® er. We show that by using di erent non-parametric spectral classi® ers to obtain the initial classi® cation, we can signi® catively improve the accuracy of the classi® cation with a reasona...
متن کاملClassi cation of Predicates and Languages
We study the classi cation of recursive predicates and languages. In particular, we compare the classi cation of predicates and languages with the classi cation of arbitrary recursive functions and with learning. Moreover, we re ne our investigations by introducing classi cation with a bounded number of mind changes and establish a new hierarchy. Furthermore, we introduce multi{classi cation an...
متن کاملClassifying Predicates and Languages
The present paper studies a particular collection of classi cation problems, i.e., the classi cation of recursive predicates and languages, for arriving at a deeper understanding of what classi cation really is. In particular, the classi cation of predicates and languages is compared with the classi cation of arbitrary recursive functions and with their learnability. The investigation undertake...
متن کاملBayesian Classi cation
Bayesian classi cation addresses the classi cation problem by learning the distribution of instances given di erent class values. We review the basic notion of Bayesian classi cation, describe in some detail the naive Bayesian classi er, and brie y discuss some extensions. C5.1.5.
متن کامل