Chaoqun Hong

Multi-view ensemble manifold regularization for 3D object recognition

View-based methods are popular in 3D object recognition. However, current methods with traditional classifiers are usually based on one-to-one view matching and fail to capture the structure information of multiple views. Some multi-view based methods take different views into consideration, but they still treat views separately. In this paper, we propose a novel 3D object recognizing method based on multi-view data fusion, called Multi-view Ensemble Manifold Regularization (MEMR). In this method, we model image features with a regularization term for SVM. To train this modified SVM, multi-view learning is achieved with alternating optimization. Hypergraph construction is used to better capture the connectivity among views. Experimental results show that the accuracy rate has been improved by 20-25%, which demonstrates the effectiveness of the proposed method.

Multi-view hypergraph learning by patch alignment framework

Graph-based methods are currently popular for dimensionality reduction. However, most of them suffer from over-simplified assumption of pairwise relationships among data. Especially for multi-view data, different relationships from different views are hard to be integrated into a single graph. In this paper, we propose a novel semi-supervised dimensionality reduction method for multi-view data. First, we assume the hyperedges in hypergraph as patches and apply hypergraph to the patch alignment framework. Second, the weights of the hyperedges are computed with statistics of distances between neighboring pairs and the patches from different views are integrated. In this way, we construct Multi-view Hypergraph Laplacian matrix and we get the dimensionality-reduced data by solving the standard eigen-decomposition to obtain the projection matrix. The experimental results demonstrate the effectiveness of the proposed method on retrieval performance.

Hypergraph-based multi-example ranking with sparse representation for transductive learning image retrieval

Content-based image retrieval (CBIR) always suffers from the so-called semantic gap. Query-By-Multiple-Examples (QBME) is introduced to bridge it and applied in a lot of CBIR systems. However, current QBME methods usually query with each example separately and combine the query results. In this way, the computational time will increase linearly with the growing number of query examples. In this paper, we propose a novel QBME method for fast image retrieval based on transductive learning framework. To improve the speed of QBME, we introduce two improvements. First, we explore the semantic correlations of image data in the training process. These correlations are learned using sparse representation. With the semantic correlations, semantic correlation hypergraph (SCHG) is constructed to model the images and their correlations. The construction of SCHG is free of any parameter. After constructing SCHG, we predict the ranking values of images by using the pre-learned semantic correlations. Second, we propose a multiple probing strategy to rank the images with multiple query examples. Different from traditional QBME methods which accept one input example at a time, all the input examples are processed at the same time in this strategy. The experimental results demonstrate the effectiveness of the proposed method on both retrieval performance and speed.

Hypergraph regularized autoencoder for image-based 3D human pose recovery

Image-based human pose recovery is usually conducted by retrieving relevant poses with image features. However, semantic gap exists for current feature extractors, which limits recovery performance. In this paper, we propose a novel feature extractor with deep learning. It is based on denoising autoencoder and improves traditional methods by adopting locality preserved restriction. To impose this restriction, we introduce manifold regularization with hypergraph Laplacian. Hypergraph Laplacian matrix is constructed with patch alignment framework. In this way, an automatic feature extractor for silhouettes is achieved. Experimental results on two datasets show that the recovery error has been reduced by 10% to 20%, which demonstrates the effectiveness of the proposed method. HighlightsPose recovery with autoencoder is imposed locality reservation with Laplacian matrix.The construction of Laplacian matrix is improved by using hypergraph optimization.

Image-Based 3D Human Pose Recovery with Locality Sensitive Sparse Retrieval

Image-based 3D human pose recovery is usually conducted by retrieving relevant poses with image features. However, it suffers from high dimensionality of image features and low efficiency of retrieving process. In this paper, we propose a novel approach to recover 3D human poses from silhouettes. This approach improves traditional methods by adopting locality sensitive sparse coding in the retrieving process. It incorporates a local similarity preserving term into the objective of sparse coding, which groups similar silhouettes to alleviate the instability of sparse codes. The experimental results demonstrate the effectiveness of the proposed method.

Realtime and robust object matching with a large number of templates

Most of conventional object matching methods are based on comparing local features, which are too computational demanding. Recently, Dominant Orientation Templates (DOT) were proposed to solve the efficiency issue. Although DOT obtains promising results, it still suffers the problem of wasting too many bits in representation and fragility when partial occlusion occurs. As the number of templates increase, the performance will decrease. Therefore, we propose a compact DOT representation with a fast partial occlusion handling approach. Instead of using seven orientations in the original implementation, we employ single orientation of the highest gradients for the proposed compact DOT representation (C-DOT). Consequently, the size of feature vectors is reduced from 8 bits to 3 bits. To efficiently tackle the partial occlusion, we introduce the C-DOT similarity map to store the matching scores of individual grids in each sliding window, which is used to further infer the occlusion map. The experimental results demonstrate that the proposed method outperforms DOT.

Semantic embedding for indoor scene recognition by weighted hypergraph learning

Conventional methods for indoor scenes classification is a challenging task due to the gaps between images? visual features and semantics. These methods do not consider the interactions among features or objects. In this paper, a novel approach is proposed to classify scenes by embedding semantic information in the weighted hypergraph learning. First, hypergraph regularization is improved by optimizing weights of hyperedges. Second, the connectivity among images is learned by statistics of objects appearing in the same image. In this way, semantic gap is narrowed. The experimental results demonstrate the effectiveness of the proposed method. A novel approach to classify scenes by embedding semantic information.A new hypergraph regularization by optimizing weights of hyperedges.Constructing connectivity among images by using statistics of objects appearing in the same image.

Anti-counterfeiting digital watermarking algorithm for printed QR barcode

Anti-counterfeiting watermarking technology provides a potential way to prevent quick response (QR) barcode from being copied or forged, and many relevant algorithms for digital QR barcode appeared recently. However, as for printed QR barcode, there is no actual research achievement published. This paper studies anti-counterfeiting techniques for printed QR barcode. The primary performance parameter, decoding rate of QR barcode as well as detection rate of digital watermark, is defined and discussed. A multi-channel robust watermarking scheme based on discrete wavelet transform (DWT) is proposed. In the watermarking scheme, the DWT domain is divided into non-overlapping rectangular areas as called watermarking channels. Channel watermark as well as anti-counterfeiting watermark is embedded into each watermarking channel. At the stage of anti-counterfeiting watermark extracting, the distortion rates of the watermarking channels are estimated by channel watermark. Based on the distortion rates, anti-counterfeiting watermark is computed by a well-designed linear estimation algorithm, whose validity is theoretically proved by analyzing anti-counterfeiting watermark bit error. The experimental results show that the detection rate of anti-counterfeiting watermark is greatly improved with the proposed scheme, and that we can correctly extract anti-counterfeiting watermark from printed QR image by choosing reasonable parameters.

Three-dimensional image-based human pose recovery with hypergraph regularized autoencoders

Three-Dimensional image-based human pose recovery tries to retrieves 3D poses with 2D image. Therefore, one of the key problem is how to represent 2D images. However, semantic gap exists for current feature extractors, which limits recovery performance. In this paper, we propose a novel feature extractor with deep neural network. It is based on denoising autoencoders and improves previous autoencoders by adopting locality preserved restriction. To impose this restriction, we introduce manifold regularization with hypergraph learning. Hypergraph Laplacian matrix is constructed with patch alignment framework. In this way, an automatic feature extractor for images is achieved. Experimental results on three datasets show that the recovery error can be reduced by 10 % to 20 %, which demonstrates the effectiveness of the proposed method.

Exemplar-based 3D human pose estimation with sparse spectral embedding

Abstract In exemplar-based approaches, human pose estimation is achieved by retrieving relevant poses with images. Therefore, image description is critical and it is common to extract multiple features to better describe the visual input data. To fuse multiple features, traditional methods simply concatenates multi-view features into a long vector. There are two shortcomings in this oversimplified process: (1) it usually results in lengthy feature vectors, which suffers from “curse of dimensionality”; (2) it is not physically meaningful and may be incapable of fully exploiting the complementary properties of multi-view features. To address such problems in this paper, we present a dimension reduction method based on sparse spectral embedding, followed by an ensemble of nearest neighbor regression in low-rank multi-view feature space, to infer 3D human poses from monocular videos. The experiments on HumanEva dataset show the effectiveness of the proposed method.