Automatic Detection of Regular Geometrical Shapes in Photograph using Machine Learning Approach

We propose a model for automatic detection of regular geometrical shapes in a photograph. The proposed framework uses a machine learning approach to detect shapes based on geometrical features. The geometrical shapes are elements of compositions in a photograph. Usually, the regular shapes play a very important role in photo aesthetic analysis. They provide a good amount of aesthetic score to photographs. The developed model is a multi-class classifier using Random Forest based on 9 distinct geometrical features, which can detect and classify the regular shapes into 'Circle', 'Rectangle', 'Square', and 'Triangle'. We test our model on a ground truth dataset containing 250 images. The experimental result shows that the proposed model gives the accuracy up to 96%, which outperforms the current state-of-the-art. Its application to the problem of aesthetic score evaluation of photographs, and online guidance to amateur photographers to improve their photography skill.