Search

Locality-constrained Linear Coding (LLC) based visual tracking can give a better and faster tracking performance than traditional sparse representation based tracking methods. However, the existing LLC based methods often use the anchor points near the target to build the sparse coding dictionary for local sparse coding. It may cause a problem that it is hard to discriminate the difference between the negative and positive anchor points in sparse coding dictionary when facing severe background clutter, illumination change and occlusion. In this paper, we propose a context aware sparse coding method to achieve robust visual tracking. The proposed method can prevent the negative anchor points from disturbing the classifier accuracy because it uses a global context regularizer to constrain the sparse coding value of those negative anchor points that are similar to the positive anchor points. Experiment results show that our method can achieve a better tracking performance than state-of-the-art tracking methods do.

Bag of Words (BOW) method with spatio-temporal local features has achieved great performance in human action recognition. However, most of the existing BOW approaches based on vector quantization (VQ) neglect the contextual information of each descriptor, and suffer serious quantization error. There are two main reasons for these: in the first, each local feature is only assigned to one label and second, the information about the spatial layout of the features is disregarded. In this paper, we present a novel and effective coding method called context and locality constrained linear coding (CLLC) to overcome these limitations, in which the relationships among local features and their structural information are preserved. After that, a group-wise sparse representation based classification (GSRC) method is implemented to assign the query sample into one category which yields the smallest reconstruction error. Our method is verified on the challenging databases and achieves comparable performance with state-of-the-art methods. We model a novel and effective human action recognition scheme.We propose context and locality constrained linear coding method to fully describe each local feature."Decisive feature" and "indecisive feature" are aided to collect the contextual information.Group-wise sparse representation based classification method is introduced to category the query samples.Our encoding problem has both analytical solution and approximation solution.

In this paper, we investigate action recognition using an inexpensive RGB-D sensor (Microsoft Kinect). First, a depth spatial-temporal descriptor is developed to extract the interested local regions in depth image. Such descriptors are very robust to the illumination and background clutter. Then the intensity spatial-temporal descriptor and the depth spatial-temporal descriptor are combined and feeded into a linear coding framework to get an effective feature vector, which can be used for action classification. Finally, extensive experiments are conducted on a publicly available RGB-D action recognition dataset and the proposed method shows promising results.

This paper discusses random linear network coding with and without the use of a Vandermonde matrix to obtain the coding coefficients. Performance comparisons of such random linear network coded networks with networks employing traditional store and forward technique are also provided. It is shown that random linear network coding using a Vandermonde matrix can improve the network utilization factor by reducing the overhead compared to random linear coding that does not use a Vandermonde matrix. Our numerical results show that random linear network coding with a Vandermonde matrix provides a considerable improvement in throughput and delay when compared to a network employing a traditional store and forward strategy. An inherent feature of random linear network coding which makes it possible to employ simple encryption techniques is also discussed.

Bag-of-Words lies at a heart of modern object category recognition systems. After descriptors are extracted from images, they are expressed as vectors representing visual word content, referred to as mid-level features. In this paper, we review a number of techniques for generating mid-level features, including two variants of Soft Assignment, Locality-constrained Linear Coding, and Sparse Coding. We also isolate the underlying properties that affect their performance. Moreover, we investigate various pooling methods that aggregate mid-level features into vectors representing images. Average pooling, Max-pooling, and a family of likelihood inspired pooling strategies are scrutinised. We demonstrate how both coding schemes and pooling methods interact with each other. We generalise the investigated pooling methods to account for the descriptor interdependence and introduce an intuitive concept of improved pooling. We also propose a coding-related improvement to increase its speed. Lastly, state-of-the-art performance in classification is demonstrated on Caltech101, Flower17, and ImageCLEF11 datasets.

Local and global features are considerably important features in computer vision and play an important role in scene categorization task. In this paper, an integrated feature description for scene categorization is constructed. First, we extract a type of extended contextual features for scene images that contain the local gradient information and more comprehensive local structural information. Mapping the local features by using improved LLC (Local-constrained Linear Coding) scheme to form the original image representation; Secondly, a set of global features named 'gist' are extracted that provide a statistical summary of the spatial layout properties of the scene; Then, the contextual features and 'gist' features are weighted combined based on their contribution for the integrated feature description, and each image is represented by using LLC scheme. Finally, we perform the scene categorization by libSVM with the HIK (Histogram Intersection Kernel) function. The proposed method achieves a satisfactory average accuracy rate 87.60% on a set of 15-scene categories.

In this paper a Walsh linear coding algorithm (WLC) is developed which yields a compressed approximation to the Walsh power spectrum of a time frame. This algorithm is interpolative, rather than predictive. It minimizes the mean square interpolative error, and it yields WLC model spectra. The effectiveness of utilizing WLC model spectra in speech recognition systems is explored. Each of two unity gain distortion measures is embedded in a dynamic time warp based speaker-dependent isolated word recognition system and is tested by means of a 10-digit vocabulary.