Object clustering by K-means algorithm with binary sketch templates
Xueyan Mei · 2016
Object clustering is a very challenging unsupervised learning problem in machine learning and pattern recognition. In this paper, we will study visual object pattern clustering problem by combining the k-means clustering algorithm and the binary sketch templates, which quantify each image by a vector of indicators showing that a sketch at certain location, scale, and orientation exist or not. This representation is very simple and accounts for shape deformation of objects by local max pooling operations. Most importantly, such representations can be visualized by meaningful symbolic sketch templates. The experiment conducting on a small clustering dataset shows that the k-means with binary sketch templates for object clustering is very promising and the learned mixture of templates is also meaningful for understanding the results.