| A single digital image, whether it is an image of handwritten characters or computer generated artwork, is inherently a static unchanging entity. Unlike video stream data or on-line character recognition data, there is no dynamic component or temporal information associated with it. Thus, even though a digital image, perhaps of handwritten characters for example, may have been created using a very specific dynamic process, information about that process is completely lost within the digital image of those characters. The unavailability of this type of dynamic information may represent a significant disadvantage for many image processing and understanding applications. One notable example here is the improved accuracy that on-line handwriting recognition systems offer versus their off-line counterparts. Here the dynamic or temporal information in addition to the binary images of the characters further aids in the recognition process due to the fact that certain characters are usually drawn in very specific ways. Hence, the extraction or reconstruction of a dynamic creation process corresponding to a specific static digital image may be very useful and is the main focus of this dissertation. The research presented here is more specifically applied to the domain of automating the generation of computerized embroidery data files. This technology is useful within the textile industry which uses computerized embroidery to imprint woven materials. Additionally, a new method for computing skeletal and distance transform information efficiently and quickly is presented. Both of these processing tasks are common components of many image understanding or recognition systems. The ability to accomplish this processing in less time on low-cost hardware may be particularly useful when near real-time speed is required within an application. |
