Muhammad Rzi Abbas Department of Mechatronics and Control Engineering
muhammadrziabbas@uet.edu.pk Lecturer, Mechatronics Dept. University of Engineering and Technology, Lahore Machine Vision System • At early stages of Machine Vision, researchers divided the Machine Vision problem into a two step problem of ‘Preprocessing’ and ‘Recognition’. • Preprocessing was referred to as tackling all generalizations • Recognition was done using Statistical Pattern Recognition techniques • Intermediate techniques, such as Hough Transforms, were never explored. Machine Vision System • Let us now generalize things. • Let us look at computer vision from an aerial view. • Loss of information in 3D to 2D conversion Machine Vision System • Let us now generalize things. • Let us look at computer vision from an aerial view. • Interpretation of Images: Past knowledge and context • Noise • Too much data Machine Vision System • Let us now generalize things. • Let us look at computer vision from an aerial view. • Local window vs need for global view: A computer sees everything through a keyhole. Even humans, sometimes, cannot interpret a scene when seeing it through a single or multiple keyholes. Machine Vision System • Image understanding by a machine can be seen as an attempt to find relation between input image(s) and previously established models of the observed world. • Transition from the input image(s) to the model reduces the information contained in the image to relevant information for the application domain. • This process is divided into several steps shown in the next slide Machine Vision System Machine Vision System • A transition from non-abstract data to highly abstract data can be seen is the bottom-up flow diagram shown in the previous slide. • This whole diagram can be sub-categorized into two main portions • Low-Level Processing: Have little knowledge about the contents of the image. Deals majorly with image compression, pre-processing methods for noise filtering, edge extraction, image sharpening etc. At this stage the image is represented as f(x,y), whose value, at simplest, is usually brightness depending on two parameters x,y, the coordinates of the location in the image Machine Vision System • A transition from non-abstract data to highly abstract data can be seen is the bottom-up flow diagram shown in the previous slide. • This whole diagram can be sub-categorized into two main portions • High-Level Processing: Based on knowledge, goals and plans of how to achieve those goals, and artificial intelligence methods. This level tries to imitate human cognition and ability to make decisions. Machine Vision System • Machine Vision Systems are expected to solve complex tasks, the goal being to obtain similar results to those provided by the biological systems. • Consider the following figure in which a particular image representation is presented. • Can you recognize this image? • Let’s have a look at a second image… Machine Vision System • So the point is that Humans use a lot of a priori knowledge, while machines begin only with arrays of numbers. • Computers are better at processing local parts of the image, while Humans see an image globally. • Hence General Knowledge, Domain-specific Knowledge, and information extracted from the image are essential components for ‘understanding’ arrays of numbers. Machine Vision System • Low-Level Processing overlaps completely with what we called Digital Image Processing. • Image Segmentation is the next step which sometimes fall into Low- Level Processing, while at others in High-Level Processing. • In this step computer tries to separate objects from each other and background. In most examples total segmentation is not possible, so partial segmentation is done. • Object description and classification in a totally segmented image are also considered as Low-Level Processing step. Machine Vision System • High-Level Processing differs from the Low-Level Processing in the data used. • LLP use images, arrays of numbers containing brightness values etc., while HLP data is the product of the LLP stage. • High-Level data represent knowledge about the image content, for example object size, shape, and mutual relationship between objects. Machine Vision System • Most Low-Level Image Processing methods were proposed in 1970s or earlier. • Recent research is trying to find more efficient and more general algorithms and trying to implement them on more sophisticated equipment, in particular, parallel machines. • An easy sounding yet complex and still unsolved problem is automating the Low-Level Image Processing!!! • Human intervention is still needed. • HLP also gives feedback to LLP but still a void remains. Machine Vision System • So much has happened but still the only solution to the ‘Vision Problem’ is our brain. Applications • Industrial Applications • Safety • Quality Control • Control of AGVs Applications • Medical Applications • PET (Positron Emission Tomography) • CAT (Computerized Axial Tomography) • MRI (Magnetic Resonance Imaging) • fMRI (functional Magnetic Resonance Imaging) • Environmental Imaging Applications • Military Applications • Detection of soldiers and/or vehicles • Missile guidance • Object recognition • UAVs • Law Enforcement and Security • Surveillance • Biometrics (Fingerprints, Iris scanner, Face recognition etc.) Applications • Consumer Electronics • Digital cameras and camcorders • Smart Phones, TVs, DVD Players • The Internet, Particularly World Wide Web • Uploading visual data • Tagging • Searching based on visual information References • Image Processing, Analysis and Machine Vision by Milan Sonka, Vaclav Hlavac and Roger Boyle, 3rd Edition, 2008. • Chapter 1 (Sections 1.2 and 1.3) • Machine Vision by David Vernon, Published in 1991 • Chapter 1 (Section 1.1) • Computer and Machine Vision – Theory, Algorithms, Practicalities by E.R.Davies, 4th Edition ELSEVIER, 2012 • Chapter 1 (Sections 1.1, 1.2.1 and 1.2.2)