MULTIMEDIA RETRIEVAL
Semester 1, 2022
Content Based Retrieval II
Visual feature extraction Shape
Copyright By PowCoder代写 加微信 powcoder
Indexing
Presentation & Evaluation
School of Computer Science
Trademark Retrieval System
Contour and skeleton strokes
Contour strokes only
Skeleton strokes only
School of Computer Science
Syntactic Representation
accaa Syntactic Representation: a c d e a c a b a
We can reduce an object to a set of structural elements, or primitives.
By adding a syntax, such as connectivity rules, it is possible to obtain a syntactic representation, which is simply a string of symbols, each representing a primitive. The syntax allows a unique representation and interpretation of the string.
The design of a syntax that transforms the symbolic and the syntactic representations back and forth is a difficult task. It requires specification of a complete and unambiguous set of rules, which have to be derived from the understanding of the scene under study.
Primitive Structural Symbols
Object Structure
School of Computer Science
Syntactic Representation
Need making use of domain knowledge
School of Computer Science
Curvature Scale Space
Curvature Scale Space (CSS)
School of Computer Science
Representations Based on CSS
CSS image
Interval tree Segment tree
School of Computer Science
http://www.ee.surrey.ac.uk/Research/VSSP/imagedb/demo.html
School of Computer Science
Invariant to affine transformation
School of Computer Science
CSS Image Matching
School of Computer Science
Interval Tree
School of Computer Science
Segment Tree
School of Computer Science
Segment Tree
Original CSS Image Traced CSS Image Aligned CSS Image
School of Computer Science
Segment Tree
School of Computer Science
Segment Tree Samples
School of Computer Science
Segment Tree
Hierarchical representation
Present scale information
Comply with human perception
Invariant to translation, rotation, and scaling Robust to noise
Potentials handling shape occlusion
School of Computer Science
Model based Shape Matching
Modal matching
Build a modal template for each prototype object and analyze
similarity between the source object and the template Elastic template matching
A shape can be regarded as a template and will be deformed in order to match with a target image.
The deformation is modeled by a discrete set of parameters.
The adaptation of these parameters is controlled through a cost
The original shape is deformed until it finds a reasonable fitting with the target image.
School of Computer Science
Modal matching
Training Samples Prototype Object
Unknown Input
Modal Matching
School of Computer Science
Elastic Template Matching
User Sketch
Database Image
School of Computer Science
Model based Shape Matching
Finding correspondence between shapes
Shape recognition/classification and retrieval,
usually for non-grid objects
To make use of domain knowledge
Computationally expensive
Refine initial retrieval results
School of Computer Science
http://www.hermitagemuseum.org/fcgi-bin/db2www/qbicLayout.mac/qbic?selLang=English
Spatial Relationship
Representation of spatial relationship
Directionalrelationship:right, left,above,…
Topological relationship: disjunction, adjacency, overlapping, …
2D String
2D G-String 2D C-String
School of Computer Science
Find centroid of each object
Project each centroid to the x- and y-axis
For (u, v) string along x- and y-axis (ABC, ACB), which is 2D string.
Minimum Boundary Rectangle
School of Computer Science
Usually, there are three spatial operators appearing in the 2D string.
< :denote the left-right relationship in u string and below-above relationship in v string.
=:denote the spatial relationship “at approximately the same spatial location as”
|:denote the edge-to-edge relationship
School of Computer Science
Complicated example
(b