Abstract
We propose an incremental spectral clustering method for stream data clustering and apply it to stream image segmentation. The main idea in our work consists of generating the data points in the kernel space by Fastfood features and iteratively calculating the eigendecomposition of data. Compared with the popular Nyström-based approximation, our work accesses each data point only once while Nyström, in particular the sampling scheme, will go through the entire dataset first and calculate the embeddings of data points with a second visit. As a result, our method is able to learn data partitions incrementally and improve eigenvector approximation with more and more data seen from a stream. By contrast, the performance of the standard Nyström is fixed when the sample set is selected. Experimental results show the superiority of our method.
Highlights
In the last decade, clustering methods are widely used in image processing and data mining, such as image segmentation [1], image matting [2], path planing [3] and thermal error modeling [4].Due to its advantages in clustering accuracy, spectral clustering plays an important role in data partition [5]
Compared with the Nyström methods, our method shows its superiority in memory usage: our method requires a constant occupation of memory, O( D2 ), where D is the dimension of Fastfood features [9] and is fixed by users
We propose an incremental image segmentation method based on Fastfood features
Summary
In the last decade, clustering methods are widely used in image processing and data mining, such as image segmentation [1], image matting [2], path planing [3] and thermal error modeling [4]. The main memory may not be sufficient Another disadvantage of Nyström for stream data is the sampling process which is essential for the approximation accuracy in Nyström. Popular sampling methods, such as the k-means or the kernel k-means sampling, need to go through the entire input data first and search for the optimal landmark points. Since most processing units in stream data clustering are memory-sensitive, it is strongly recommended to avoid any storage of history data This clustering method is expected to obtain the final partition of the entire dataset according to batch-clustering-results.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.