Variable Size Blob Detection with Feature Stability

This paper proposes an object detection method using the methods of variable size blob detection along with feature stability. The objective of this algorithm is for detecting variable size and variable shape objects without the use in preceding information of the object of interest. Process of the algorithm constructs a scalespace tree from blobs detected from a series of images after blurring. Features and spatial information of the blob provides a role in constructing trees. Nodes that are overlapped from consecutive scales will be linked to form a branch only if the features between them are stable. Standard deviation and average gray level were used to determine its robustness. Feature vector can be extended or modified to suit several applications involving medical purposes. Blob features such as compactness can also be added for specific applications such as variable-size blood cell detection. Variable size blob detection with feature stability is compared with other 2 conventional methods in order to demonstrate its performance. This algorithm can be used as a preprocessing step in an object or a pattern recognition application.