Detection and classification of human teeth in photographs using convolution mask and the watershed algorithm

This research presents a novel image-processing algorithm for classifying human teeth in digital photographs, utilizing a combination of convolution masks and the watershed segmentation algorithm. The primary objective is to accurately identify and classify different types of lower teeth in simulated images derived from the SimKit model, a standardized dental simulation framework. The study focuses on digital images of lower teeth captured using a conventional digital camera, simulating real-world photographic conditions. These images are preprocessed through a series of color space transformations and morphological operations designed to enhance the visibility and separation of dental structures from surrounding elements, such as gums or background noise. Since previous studies on photographic images of human teeth have been limited, this work addresses the gap by improving posterior-tooth detection, where conventional watershed methods are less effective. Building on prior use of features such as cusps, grooves, and ridges, the study further employed convolution masks to detect posterior teeth, while watershed segmentation remained effective for the anterior teeth. To implement the classification system, custom software was developed using MATLAB R2020b. This software applies convolution masks to enhance image features, followed by the watershed algorithm, which segments individual teeth and facilitates their classification based on morphological characteristics. The performance of the classification algorithm was quantitatively evaluated using error rate ratios, with the benchmark set at an acceptable error ratio not exceeding 1.00. The evaluation results indicate promising classification accuracy across different tooth categories: incisors exhibited an error ratio of less than or equal to 0.12; canines showed an error ratio of less than or equal to 0.36; premolars maintained a ratio of up to 1.00; and molars demonstrated error ratios not exceeding 0.75. The results confirm that the proposed method is capable of effectively identifying lower teeth in photographic images with high accuracy. This research contributes to the development of automated dental analysis systems and supports the efficient creation of comprehensive databases of tooth types derived from oral photographs, potentially aiding both clinical diagnostics and educational tools in dentistry.