China 
Africa 
Explore chapters and articles related to this topic
Multiscale Data Condensation
Published in Sankar K. Pal, Pabitra Mitra, Pattern Recognition Algorithms for Data Mining, 2004
The method for data condensation, discussed in Section 2.1, obtains condensed sets of different degrees of detail by varying a scale parameter k. It may be noted that such variable detail representation may be achieved by other approaches also, including random sampling. However, unlike random sampling the scales induced by the density-based multiscale method are not prespecified by the sizes of the condensed sets but follow the natural characteristics of the data. As far as efficiency of the scaling procedure is concerned, it may be noted that in most of the multiscale schemes for representing data or signal, including wavelets, efficiency is achieved by a lenient representation of the ‘unimportant’ regions and a detailed representation of the ‘important’ regions, where the notion of importance may vary from problem to problem. The condensation algorithm follows a similar principle where at each scale the different regions of the feature space are represented in the condensed set based on the densities of those regions estimated at that particular scale. Figure 2.2 illustrates the concept of variable scale representation. The data consists of 2000 points selected randomly from two nonoverlapping circles of radius 1 unit and centers at (2,0) and (5,0) respectively (Figure 2.2(a)). Figures 2.2(b) —(e) shows representation of the data by condensed sets at different levels of detail. It can be seen that in Figure 2.2(b) only two points cover the entire data set. In Figure 2.2(c) four points are used to represent the entire data set. Figure 2.2(d) and (e) are more detailed representations of the data.
Estimating signal-dependent noise (SDN)-based motion variations to enhance gesture recognition
Published in Advanced Robotics, 2022
Swagata Das, Yuya Ishibashi, Mayuko Minakata, Yuichi Kurita
The condensation algorithm that was initially developed to track rapid movement of objects was also applied to gesture recognition. Facial and hand gesture recognition was attempted using an extended condensation algorithm that applies random sampling for incremental matching of estimated human motion trajectories and input data [16]. A multi-modal gesture recognizer with condensation algorithm was proposed to identify human gestures and facial expressions by deploying decision and feature-based fusion strategies. This work reported an accuracy of 92.6% and proved that identifying facial expressions may help improve the accuracy of recognizing ambiguous hand gestures [17]. Condensation algorithm was, however, not popular during the recent years in the field of gesture recognition as it was outperformed by modern algorithms.