A summary of image segmentation techniques

摘要: Machine vision systems are often considered to be composed of two subsystems: low-level and high-level vision. Low level consists primarily image processing operations performed on the input produce another with more favorable characteristics. These may yield images reduced noise or cause certain features emphasized (such as edges). High-level includes object recognition and, at highest level, scene interpretation. The bridge between these subsystems is segmentation system. Through segmentation, enhanced mapped into a description involving regions common which can used by higher tasks. There no theory segmentation. Instead, techniques basically ad hoc differ mostly in way they emphasize one desired properties an ideal segmenter balance compromise property against another. categorized number different groups including local vs. global, parallel sequential, contextual noncontextual, interactive automatic. In this paper, we categorize schemes three main groups: pixel-based, edge-based, region-based. Pixel-based classify pixels based solely their gray levels. Edge-based first detect discontinuities (edges) then use that information separate regions. Finally, region-based start seed pixel (or group pixels) grow split until original only homogeneous Because there survey papers available, will not discuss all schemes. Rather than survey, take approach detailed overview. We focus approaches order give reader flavor for variety available yet present enough details facilitate implementation experimentation.