Cluster Based Approaches for Keyframe Selection in Natural Flower Videos

The selection of representative keyframes from a natural flower video is an important task in archival and retrieval of flower videos. In this paper, we propose an algorithmic model for automatic selection of keyframes from a natural flower video. The proposed model consists of two alternative methods for keyframe selection. In the first method, K-means clustering is applied to the frames of a given video using color, gradient, texture and entropy features. Then the cluster centroids are considered to be the keyframes. In the second method, the frames are initially clustered through Gaussian Mixture Model (GMM) using entropy features and the K-means clustering is applied on the resultant clusters to obtain keyframes. Among the two different sets of keyframes generated by two alternative methods, the one with a high fidelity value is chosen as the final set of keyframes for the video. Experimentation has been conducted on our own dataset. It is observed that the proposed model is efficient in generating all possible keyframes of a given flower video.