Digital Image Processing Introduction
Course Info Course Outline Course website http://cs.gmu.edu/~asood/cs686/cs686-sood-a.html Course website http://cs.gmu.edu/~asood/cs686 9/18/2018
Overall Motivation 1 Improvement of pictorial information for human interpretation Improve digitized newspaper pictures Circa 1920: 5 gray levels; 1929 : 15 levels 1960s correct for distortions introduced by on-board cameras Modern – X-rays, pollution, pattern recognition, art work, micro-arrays, sharing of images (collaboration) Panoramic views 9/18/2018
Overall Motivation 2 Processing of scene data for autonomous machine perception (Machine Visual Perception) Automatic Character Recognition Courtesy Amount Reading in checks – what error rate is acceptable? Industrial robots, screening of x-rays and blood samples, crop assessment Object Recognition, Planning and Communication 9/18/2018
Overall Motivation 3 Build systems that use imaging as an enabling technology Remote desk top Street location recognition Parking lot management GIS 9/18/2018
System Overview Display Hard copy Transmission Remote Processing & Storage Display Hard copy Transmission Remote Processing & Object/ scene Image Capture (Acquisition) CPU User Interaction Image Sensor A/D Conversion Frame Processor IP Workstation Zoom Scroll IP functions (MATLAB) Storage Requirements Large Processing: Real time 60 frames per sec/ On-line/Off-line Enhancements -Parallel -Pipeline 9/18/2018
Sensors What is measured? Visual – intensity: Luminance of object in the scene Thermal – temperature: infra red Xrays – absorption characteristics Ultrasonic scanning Laser scanners: 3-D images Radars Magnetometers Gravity meters 9/18/2018
Assessing Sensor Quality Resolution Uniformity of grid 2-D or 3-D images Indirect measurement Noise effects 9/18/2018
Applications Active vs Passive Remote sensing via satellite Agriculture monitoring Land use Weather Flood and fire control Defense intelligence Environment monitoring 9/18/2018
Business / industry Scanning Re-use Multiple locations Security Fax Robots Industry, defense, consumer, environment Medical Patient screening and monitoring, treatment planning 9/18/2018
Overall tasks of IP & CV Object Recognition Planning Communications Unmanned vehicles / intelligent robotics: Perception, planning and action cycle Architecture Sensors Algorithms 9/18/2018
IP Problems Image Representation and Modeling Image Enhancement Image Restoration Image Analysis Image Reconstruction Image Compression 9/18/2018
Image Representation and Modeling Fidelity or intelligibility criteria Design and evaluation of imaging sensor Sampling of a BW TV signal Models of perception Contrast, color, spatial frequencies Sampling rate, quantization levels and errors Represent images as a combination of basic images Characterization of local behaviors 9/18/2018
Image Representation and Modelling Perception Models Fidelity Temporal perception Scene perception Local models Image quantization Deterministic (transforms) Statistical (time series) Global AI / Scene analysis Sequential and clustering Image understanding 9/18/2018
Fidelity of Image Sampling How to assess fidelity? Often based on quality measures. Reconstruction image from sampled image. Black and white TV signal has about 4 MHz bandwidth. What rate should it be sampled? Sampling rate (Nyquist) > 8,000,000 per sec Frames samples in 1/30 sec Samples per frame = 8000000/30 = 266,000 No of lines per frame = 525 No of samples per line = 266,000/525=500 If 512 lines per frame, then 512 samples / line 9/18/2018
Image Enhancement Reduce noise Accentuate certain image features Techniques Contrast enhancement Edge enhancement Noise filtering Sharpening Magnifying Methods are usually iterative and application dependent Picture of mars, X-ray 9/18/2018
Image Restoration - 1 Objective is to minimize or remove known degradations in the image. Sensor induced – noise, geometric distortions non-linearities Camera calibration Given Image and Sensor Transfer Function estimate the object 9/18/2018
Image Restoration - 2 Transfer Function Object f(a,b) Sensor h(x,y;a,b) Image g(x,y) Objective: Given blurred image g(x,y), PSF{h(,;,)} and noise characteristics Find f(a,b) 9/18/2018
Enhancement – Restoration Similar objective Enhancement – more heuristic Restoration – mathematical model driven 9/18/2018
Image Reconstruction from Projections Projections to 3-D rendering CT scanners 9/18/2018
Image Encoding and Compression Image potentially require large storage – maybe GBs 1K x 1K x 12 bits/pixel requires 1.5MB Can we reduce the number of bits per pixel? Impact on quality Fidelity Lossless vs lossy Applications in telemedicine, videoconferencing, … 9/18/2018
Image Analysis Making quantitative measurements from images Input to the object recognition, planning tasks Often relies on segmentation Isolates objects 9/18/2018
Image Segmentation Edge detection Region growing Occlusion, overlapping objects Scale and rotation 9/18/2018