Lu Y.-SHua S.-C.JIAN-JIUN DING2022-04-252022-04-252021https://www.scopus.com/inward/record.uri?eid=2-s2.0-85124163023&doi=10.1109%2fISPACS51563.2021.9650990&partnerID=40&md5=e092595fadf40082d634e7eb4b126129https://scholars.lib.ntu.edu.tw/handle/123456789/607200Context modeling is to classify the input data into several classes and apply different probability models for each class. It plays a critical role in data compression. Conventionally, when constructing the context, one often uses several criterions to separate each axis in the feature domain into several parts independently. However, it does not consider the correlation among different features. In this work, we propose an advanced context modeling method based on over-segmentation and merging in the feature domain using the k-means clustering method. It considers the relation among different features and the number of training data corresponding to each context is more balanced. The experiments in image AC term encoding show that, with the proposed context assignment method, the coding efficiency can be much improved. ? 2021 IEEE.Adaptive arithmetic codingContext modelK-means clusteringEncoding (symbols)Image codingImage enhancementImage segmentationSignal encodingContext modelsEncodingsFeature domainInput datasK-means clustering methodK-means++ clusteringModel methodOver segmentation and mergingProbability modellingconference paper10.1109/ISPACS51563.2021.96509902-s2.0-85124163023