An Interactive Category Labeling Method for Sketch Dataset

Abstract

The invention discloses an interactive category labeling method for a sketch data set in a computer, comprising the following steps: a learning process, performing multi-feature extraction on the marked sketch data set, feature space metric learning, and calculating a distance metric function. The selection process, if it is judged that there is no unlabeled sketch in the sketch data set, it will end and the final result will be obtained. Otherwise, according to the result of metric learning, construct the feature space of the sketch dataset to be labeled, perform hierarchical clustering, and select the optimal sample subset. Online labeling, the user interactively confirms the sketches in the optimal sample subset, labels the confirmed samples, and updates the labeled sketch dataset. The remaining non-similar sketches will remain unlabeled, and the dataset of sketches to be labeled will be updated. Then, the above process is repeated continuously until the user completes all sketch annotations and obtains the final annotation result.

Description

technical field [0001] The invention relates to a processing method of a visual data set, which belongs to the technical field of computer vision, in particular to an interactive category labeling method for a sketch data set in a computer. Background technique [0002] As one of the oldest communication methods of human beings, sketching is the basic way for human beings to abstract and conceptualize visual information. In recent years, with the popularization of touch interactive devices such as smartphones and tablets, sketching has become one of the most common ways of human-computer interaction, and has been used in image and model retrieval, such as literature 1: M.Eitz, K.Hildebrand, T. Boubekeur and M. Alexa. Sketch-based image retrieval: Benchmark and bag-of-features descriptors. IEEE Transactions on Visualization and Computer Graphics, vol.17, no.11, pp.1624–1636, 2011., Literature 2: M .Eitz, R.Richter, T.Boubekeur, K.Hildebrand and M.Alexa. Sketch-based shape re...

AI Technical Summary

Problems solved by technology

[0005] To sum up, when labeling the sketch dataset as a whole, the existing supervised sketch classification methods need to predict the category and sample training, which cannot be applied to the situation where the label category is unknown or uncertain; while the unsupervised image category discovery method There will be problems in the following three aspects: 1) the method of "labeling all at once" is adopted, which relies on the classification effect of unsupervised learning, and it is difficult to guarantee the correctness of the labeling results; 2) only the Euclidean distance calculation between the underlying features of the sample is used Similarity, ignoring the category information provided by the user in the labeling process, it is difficult to fully reflect the user's interaction intention; 3) When providing category labels for the clusters to be labeled, there is a lack of reasonable screening strategies to effectively reduce the labeling burden

Images