| Object recognition is the process of identifying an object from its image. Before any identification process can begin the color image must be processed. By processing the color image of an object, different areas would be separated from each other. Features of the isolated areas can then be extracted and compared to the learned objects in the system data base and the best match found. Object recognition has applications in robotics vision, obstacle and goal identifications, product classification, industrial quality control, and visual inspection and diagnosis.; After a digital color image is grabbed or scanned, it must be smoothed to remove external noises, especially the ones associated with the camera or scanner. Edge detection is performed on the smoothed image to separate areas or segments from each other. The edge information is used in conjunction with the smoothed image for segmentation. In the segmented image, each segment belongs to only one object and has an enclosed contour. Features are then extracted and segment contours coded. The best fitting straight or curve lines are then found. Objects in the data base are then compared to the extracted lines and features using graph theory and the best match is found. Fuzzy operators and thresholds were used throughout this work to propagate uncertainties and delay decision makings for later stages. |
