Computer Vision Methods

Garant předmětu:

Lecturer:

Tutor:

Supervisor:

Department of Cybernetics

Synopsis:

The course covers selected computer vision problems: search for correspondences between images via interest point detection, description and matching, image stitching, detection, recognition and segmentation of objects in images and videos, image retrieval from large databases and tracking of objects in video sequences.

Requirements:

knowledge of calculus and linear algebra.

Syllabus of lectures:

1.Introduction. Course map. Overview of covered problems and application areas.

2.Detectors of interest points and distinguished regions. Harris interest point (corner) detector, Laplace detector and its fast approximation as Difference of Gaussians, maximally stable extremal regions (MSER).Descriptions of algorithms, analysis of their robustness to geometric and photometric transformations of the image.

3.Descriptors of interest regions. The local reference frame method for geometrically invariant description. The SIFT (scale invariant feature transform) descriptor, local binary patterns (LBP).

4.Detection of geometric primitives, Hough transfrom. RANSAC (Random Sample and Consensus).

5.Segmentation I. Image as a Markov random field (MRF). Algorithms formulating segmentation as a min-cut problem in a graph.

6.Segmentation II. Level set methods.

7.Inpainting. Semi-automatic simple replacement of a content of an image region without any visible artifacts.

8.Object detection by the „scanning window“ method, the Viola-Jones approach.

9. Using local invariant description for object recognition and correspondence search.

10.Tracking I. KLT tracker, Harris and correlation.

11.Tracking II. Mean-shift, condensation.

12.Image Retrieval I. Image descriptors for large databases.

13.Image Retrieval II: Search in large databases, idexation, geometric verification

14.Reserve

Syllabus of tutorials:

1. - 5. Image stitching. Given a set of images with some overlap, automatically find corresponding points and estimate the geometric transformation between images. Create a single panoramic image by adjusting intensities of individual images and by stitching them into a single frame.

6. - 9. Segmentation and impainting. Implement a simple impainting method, i.e. a method allowing semi-automatic simple replacement of a content of an image region without any visible artifacts.

7. -12. Detection of a instance of a class of objects (faces, cars, etc.) using the scanning window approach (Viola-Jones type detector).

13.-14. Submission and review of reports.

Study Objective:

The methods for image registration, retrieval and for

object detection and tracking are explained. In the labs,

the students implement selected methods and test

performance on real-world problems.

Study materials:

1.M. Sonka, V. Hlavac, R. Boyle. Image Processing, Analysis and Machine Vision. Thomson 2007

2.D. A. Forsyth, J. Ponce. Computer Vision: A Modern Approach. Prentice Hall 2003

Note:

Further information:

No time-table has been prepared for this course

The course is a part of the following study plans: