Image segmentation has been exhaustively researched in the past 20 years. Recently, some 3D mesh segmentation algorithms were introduced.
Image segmentation has been exhaustively researched in the past 20 years. Recently, some 3D mesh segmentation algorithms were introduced. However, no known algorithm has ever used both domains to produce a “compatible segmentation”, which might prove to be an important link between the two domains. Compatible segmentation can be used in various computer graphics applications such as texture mapping and animations as well as computer vision applications.
Segmenting an image requires human guidance. The process of helping the application\algorithm is sometimes complex and annoying. The user operation requires him to mark each segment at least once and give a very precise indication about the segments. We want to create a process that will minimize the user’s operations and the errors (incorrect segmentation).
Using a segmented 3D model that is similar to the object in the image, we fix the model to match the image as good as we can. Then, the segmented model is used in the algorithm to add pixels from the image to each segment. The process is iterative and each iteration the segment gets “bigger”.
The application is written in Matlab, with a GUI written in Matlab also (guide). The algorithm is written in Matlab.
The segmentation problem is a major bottleneck in image processing, especially in automated recognition. There is no full automatic process that can segment an image, and even those which are user guided, do not give perfect results. We have tried 2 different solutions from 2 different angles of approach. We have learned that our parameters for ascribing pixels to segments are not sufficient and more aspects should be taken into consideration.
We want to give a warm thanks to Sagi Katz, our project supervisor, who has helped us during the entire project, supported us with coding, professional literature and good ideas. We are also grateful to the lab’s stuff and to the Ollendorff Minerva Center Fund who supported this project.