Active Shape Models with Template Matching
for Automatic Incisive-Teeth Selection and Extraction

Automatic selection of areas in images is a very useful tool in Machine Learning, as these automatically selected pixels can constitute inputs in further stage model. In many occasions not all areas of an image are of interest and their use increases the dimensionality of the input, which is problematic on itself due to the Curse of Dimensionality. One would like to cut out this needless information. Usually, in large training sets, manually selecting the areas of interest is impractical. ACTIVE SHAPE MODELS with TEMPLATE MATCHING solves this problem.

Challenge: Given a training set of panoramic dental X-Rays, a program must be designed and trained to find and extract the 4 upper incisor teeth automatically. Selecting the pixels of the image that correspond to the target incisors -as mentioned before-, can be used as input for a trainable model on a further stage. However, this possible next stage is out of the scope of this particular project.


Example of a training image. Source: Wikipedia. Author: DRosenbach

Solution: It was selected the method Active Shape Models (ASM) in conjunction with Template Matching (TM) for obtaining an initial approximate position - requirement of ASM.

Programing Language and Packages: It was a requirement for this project to use Python. Additionally, it was used OpenCV for image manipulation, NumPy for mathematical computations and array manipulation, and SciPy, particularly the Fourier Transform Pack: FFTpack.