Wei Baogang

Cubic algebraic curves based on geometric constraints

Methods for curve modeling with cubic algebraic curves based on geometric constraints are introduced in this paper. A 1-parameter family of cubic curves with four given points as well as two tangent lines at the endpoints is constructed in the first part of the paper. A 1-parameter family and 2-parameter family of cubic curve constructions are presented for interpolating two given endpoints and two given tangent lines as well as two given curvatures at the endpoints in the last part of the paper.

Chinese calligraphic style representation for recognition

Chinese calligraphy draws a lot of attention for its beauty and elegance. The various styles of calligraphic characters make calligraphy even more charming. But it is not always easy to recognize the calligraphic style correctly, especially for beginners. In this paper, an automatic character styles representation for recognition method is proposed. Three kinds of features are extracted to represent the calligraphic characters. Two of them are typical hand-designed features: the global feature, GIST and the local feature, scale invariant feature transform. The left one is deep feature which is extracted by a deep convolutional neural network (CNN). The state-of-the-art classifier modified quadratic discriminant function was employed to perform recognition. We evaluated our method on two calligraphic character datasets, the unconstraint real-world calligraphic character dataset (CCD) and SCL (the standard calligraphic character library). And we also compare MQDF with other two classifiers, support vector machine and neural network, to perform recognition. In our experiments, all three kinds of feature are evaluated with all three classifiers, respectively, finding that deep feature is the best feature for calligraphic style recognition. We also fine-tune the deep CNN (alex-net) in Krizhevsky et al. (Advances in Neural Information Processing Systems, pp. 1097–1105, 2012) to perform calligraphic style recognition. It turns out our method achieves about equal accuracy comparing with the fine-tuned alex-net but with much less training time. Furthermore, the algorithm style discrimination evaluation is developed to evaluate the discriminative style quantitatively.

Fast image-based Chinese calligraphic character retrieval on large scale data

Chinese calligraphy is the art of handwriting, it draws a lot of attention for its beauty and elegance. In CADAL, a Calligraphic Character Dictionary (CCD) which contains hundreds of thousands of character images labeled with semantic meaning has been constructed and provided online to common users. It is a great challenge to perform quick and accurate image-based calligraphic character retrieval on CCD. In this paper, a novel shape descriptor, Oriented Shape Context (OSC) is proposed basing on the tranditional Shape Context (SC) to perform similarity searching. Together with GIST, GIST-OSC descriptor is proposed to represent calligraphic character image for efficient and effective retrieval. In addition, an effective retrieval schema is proposed. The retrieval schema works in two steps. Firstly approximate nearest neighbors of the query image are found quickly using GIST and then one-to-one fine matching between approximate nearest neighbors and the query image is performed using OSC. Our experiments show that the GIST-OSC descriptor and the retrieval schema are efficient and effective for Chinese calligraphic character retrieval on large scale data.

Msuggest: a semantic recommender framework for traditional chinese medicine book search engine

Learning traditional Chinese medicine knowledge from the digital library is becoming more and more important these days in China. In medicine learning, many readers want to find out the intrinsic relation between two medicines or among thousands of medicines. A semantic recommender system is useful for readers to understand something quickly by means of analogy which is a cognitive process of transferring information from a particular subject to another if they are similar in some aspects. In view of these above, we present a novel recommender framework called Msuggest to give the diverse semantic recommended medicine terminologies and book pages when a reader searching for medicine information in digital library. Users can choose various aspects including medicine property, efficacy, clinical application, place of origin, book provenance and etc. to see different recommended results. We evaluate Msuggest under the t-test on the samples from random sampling. The result shows that Msuggest is effective and efficient in giving the recommended words and book pages.

Range image registration using hierarchical segmentation and clustering

An accurate, robust, and automatic registration of overlapping range images is usually a pre-requisite step for range image analysis and applications. While accurate depiction of object geometry requires the increase of the resolutions of images and thus, the amount of data to process, an efficient processing of such data then usually becomes an issue. In this paper, we first employ the efficient tensor analysis and k means clustering methods to hierarchically segment and cluster the original range images into a small number of planar patches represented as the closest points in the original images to their centroids. Then an advanced ICP variant is adopted to register such closest points. Finally, another ICP variant is used to refine the registration results obtained over all the points in the images. The experimental results based on real range images show that the proposed technique significantly outperforms the selected two state of the art ones for accurate and efficient registration of overlapping range images.An accurate, robust, and automatic registration of overlapping range images is usually a pre-requisite step for range image analysis and applications. While accurate depiction of object geometry requires the increase of the resolutions of images and thus, the amount of data to process, an efficient processing of such data then usually becomes an issue. In this paper, we first employ the efficient tensor analysis and k means clustering methods to hierarchically segment and cluster the original range images into a small number of planar patches represented as the closest points in the original images to their centroids. Then an advanced ICP variant is adopted to register such closest points. Finally, another ICP variant is used to refine the registration results obtained over all the points in the images. The experimental results based on real range images show that the proposed technique significantly outperforms the selected two state of the art ones for accurate and efficient registration of overlapping range images.