CIS 601 Fall 2004 Introduction to Computer Vision

CIS 601 Fall 2004 Introduction to Computer Vision and Intelligent Systems Longin Jan Latecki Parts are based on lectures of Rolf Lakaemper and David Young

Computer Vision ?

Computer Vision ? “Computer vision’s great trick is extracting descriptions of the world from pictures or sequences of pictures” (Forsyth/Ponce: Computer Vision)

Pictures/Movies: How to • Represent • Process / Prepare • Handle • Recognize Objects

Representation • Digital Images • Color Spaces • Gray Images • Binary Images • Geometrical Properties

Representation • Digital Images • Color Spaces • Gray Images • Binary Images • Geometrical Properties

How to process / prepare: • Filters • Edges • Geometric Primitives • Lines, Circles

Introduction to Image Analysis and Processing

Low Level Object Handling: • Image / Video Compression • Huffman • JPEG • MPEG • …

Image File Formats JPEG - Joint Photographic Experts Group JPEG is designed with photographs in mind. It is capable of handling all of the colors needed. JPEGs have a lossy way of compressing images. At a low compression value, this is largely noticeable, but at high compression, an image can become blurry and messy. BMP - Bitmap Format uses a pixel map which contains line by line information. It is a very common format, as it got its start in Windows. This format can cause an image to be super large.

GIF - Graphics Interchange Format GIF is the most popular on the Internet, mainly because of its small file size. It is ideal for small navigational icons and simple diagrams and illustrations where accuracy is required, or graphics with large blocks of a single color. The format is loss-less, meaning it does not get blurry or messy. The 256 color maximum is sometimes tight, and so it has the option to dither, which means create the needed color by mixing two or more available colors. GIF use a simple technique called LZW compression to reduce the file sizes of images by finding repeated patterns, but this compression never degrades the image quality. GIF can also be animated.

Low Level Object Handling: • Object representation

Low Level Object Handling: • Segmentation

The “bottom-up” approach These operations fit into a processing scheme strongly associated with David Marr, whose seminal book Vision appeared in 1980. Marr espoused a principle of least commitment, and proposed a processing scheme involving a series of representations: • Grey level array (the image, in effect) • Raw primal sketch (edges) • Primal sketch (groupings of edges) • Two-and-a-half-D sketch (surface depths and orientations, camera centered) • 3 -D model (object-centered shapes and relationships). In some sense, the 3 -D model is taken as the goal of the visual processing. It can be used for matching against a database of object shapes to achieve object identification.

But that is not the whole story A better goal is to produce systems that enable successful interaction with the environment. Interaction may mean, for example: • navigating a robot or autonomous vehicle through obstacles, or along a road; • moving a robot arm to manipulate parts for assembly; • recognizing human gestures and movements for computer control; • identifying images in a database on the basis of their content.

• For many applications, a top-down, model-based or hypothesis-driven approach is more successful. In such an approach the system starts from an assumption about what is in front of it, and tests and updates this hypothesis to attempt to match the image data. • Vision is becoming increasingly dynamic. Change and motion are integral to the goals and methods, not simply techniques for recognizing shape or inferring the third dimension. Dynamic vision needs to be predictive and goal-directed. • Biological vision remains the most important inspiration for computer vision. Increasing attention is being paid to the role of foveal vision and eye movements. And computer modeling continues to shed light on how biological visual systems work.

Object Recognition: • Color, Texture, Shape

Object Recognition: • Applications • • • Character recognition Face Recognition Shape Recognition (Image Databases)

3 D Distance Histogram

The Interface (JAVA – Applet)

The Sketchpad: Query by Shape

The First Guess: Different Shape - Classes

Selected shape defines query by shape – class

Result

Intelligent Systems?

New sensors are about to make a breakthrough in development of intelligent systems Photo from NIST Laser Detection and Ranging (LADAR) is about to become the 3 D spatial sensing tool in many disciplines.

Swiss Ranger from CSEM

Intelligent Systems in Close Future • Rescue robots: http: //www. isd. cme. nist. gov/projects/USAR/ • Autonomous transport vehicles • Intelligent surveillance video analysis • Visual control devices • Image retrieval software