Image Processing and Object Detection
February 2009 ~ June 2010
Research Assistant, Computer Graphics & Vision Processing Lab
A state key project supported by the National Natural Science Foundation of China under Grant No. 60973085
Supervisor: Chuanbo Chen ([email protected])
(A) Participates in deriving a NAM with the concept of packing problem; NAM has less limitation when representing an image for its non-symmetry hierarchy, thus it can acquire higher compression ratio and do some image processing directly and more quickly.
The Non-symmetry and Anti-packing pattern representation Model (NAM) is an anti-packing problem. The idea of the NAM can be described as follows: Giving a packed pattern (a packed container) and n predefined subpatterns (n predefined objects) with different shapes, pick up these subpatterns (objects) from the packed pattern (the packed container) and then represent the packed pattern (the packed container) with the combination of these subpatterns (objects). The procedure of the non-distortion coding can be obtained by the following expression:
The Non-symmetry and Anti-packing pattern representation Model (NAM) is an anti-packing problem. The idea of the NAM can be described as follows: Giving a packed pattern (a packed container) and n predefined subpatterns (n predefined objects) with different shapes, pick up these subpatterns (objects) from the packed pattern (the packed container) and then represent the packed pattern (the packed container) with the combination of these subpatterns (objects). The procedure of the non-distortion coding can be obtained by the following expression:
where Γˊis the reconstruction pattern; P = { p1, p2, …, pn } is a set of some predefined subpatterns; n is the type number of the subpatterns; pj is the jth subpattern (1 <= j <= n) ; v is the value of pj ; A is a parameters set of the subpattern pj ; ai(1<=i<=mi) is a parameters set of shapes forpj ; m is the serial number of pj ; ε(d) is a residue pattern; and d is a threshold of ε(d) .
(B) Uses the NAM for object detection in images and videos.
Main responsibilities include:
Ⅰ. the collection of widely used datasets
Ⅱ. selection of edge direction histogram (EDH) for object detection
Ⅲ. training of an SVM classifier
Ⅳ. data analysis and processing of object detection
Ⅴ. conducts comparative studies between our method with other methods
Our approach has two methods to deal with the variation of object, both global and local.
Firstly, we propose a non-symmetry and anti-packing objectpattern representation model (NAM) to represent an object category.
Secondly, the descriptors of sub-pattern can deal with the local variation of object.
Global Spatial relation: The spatial relations between top level and thesecond level can be described by global spatial structure.
Local feature encoding: Between second level and third level, we capture different sub-pattern cues from the sub-pattern window, each type of cue is encoded by using an appropriate descriptor.
Local feature encoding: Between second level and third level, we capture different sub-pattern cues from the sub-pattern window, each type of cue is encoded by using an appropriate descriptor.
_I use shape information to represent the sub-patterns and edge direction histogram (EDH) to describe the shape information.
_The pipeline of our detection framework as follow:
first, training a classifier for each sub-pattern.
Next, using one classifier to detect hypothesis of object location, i.e., initial detection.
After that, a verification scheme is applied to the hypothesis to obtain final detection.
first, training a classifier for each sub-pattern.
Next, using one classifier to detect hypothesis of object location, i.e., initial detection.
After that, a verification scheme is applied to the hypothesis to obtain final detection.
(C) Conducts experimental data analysis of rectangle NAM, triangle NAM, and Euler number computing with the model.
(D) Participates in research on a novel NAM-based representation algorithm for color images.
(E) Analyzes the time and space complexity of a trapezium-based NAM representation method for binary images.
(F) Develops NAM image experimental system for future applications.
Publication
_Chuanbo Chen, Guangwei Wang, and Xiaochen Wang,
“Using Non-symmetry and Anti-packing Representation Model for Object Detection” , submitted for publication inJournal of Zhejiang University-SCIENCE C (Computers & Electronics)(SCI), 2010. (National Natural Science Foundation of China Grant No. 60973085)
“Using Non-symmetry and Anti-packing Representation Model for Object Detection” , submitted for publication inJournal of Zhejiang University-SCIENCE C (Computers & Electronics)(SCI), 2010. (National Natural Science Foundation of China Grant No. 60973085)
paper.pdf | |
File Size: | 344 kb |
File Type: |
code.pdf | |
File Size: | 190 kb |
File Type: |