Accuracy of PTD and PO is compared with the exact solutions for a plane wave diffraction from a large conducting disk, ten wavelength in diameter. The excellent accuracy of PTD is demonstrated, while serious errors of PO in polarization prediction are pointed out. The multiple diffraction turned out to be the main reason for the PTD error.

In this paper, we address the problem of non-parametric template matching which does not assume any specific deformation models. In real-world matching scenarios, deformation between a template and a matching result usually appears to be non-rigid and non-linear. We propose a novel approach called local rigidity constraints (LRC). LRC is built based on an assumption that the local rigidity, which is referred to as structural persistence between image patches, can help the algorithm to achieve better performance. A spatial relation test is proposed to weight the rigidity between two image patches. When estimating visual similarity under an unconstrained environment, high-level similarity (e.g. with complex geometry transformations) can then be estimated by investigating the number of LRC. In the searching step, exhaustive matching is possible because of the simplicity of the algorithm. Global maximum is given out as the final matching result. To evaluate our method, we carry out a comprehensive comparison on a publicly available benchmark and show that our method can outperform the state-of-the-art method.

In this paper, we address the problem of projective template matching which aims to estimate parameters of projective transformation. Although homography can be estimated by combining key-point-based local features and RANSAC, it can hardly be solved with feature-less images or high outlier rate images. Estimating the projective transformation remains a difficult problem due to high-dimensionality and strong non-convexity. Our approach is to quantize the parameters of projective transformation with binary finite field and search for an appropriate solution as the final result over the discrete sampling set. The benefit is that we can avoid searching among a huge amount of potential candidates. Furthermore, in order to approximate the global optimum more efficiently, we develop a level-wise adaptive sampling (LAS) method under genetic algorithm framework. With LAS, the individuals are uniformly selected from each fitness level and the elite solution finally converges to the global optimum. In the experiment, we compare our method against the popular projective solution and systematically analyse our method. The result shows that our method can provide convincing performance and holds wider application scope.

Tracking algorithms for arbitrary objects are widely researched in the field of computer vision. At the beginning, an initialized bounding box is given as the input. After that, the algorithms are required to track the objective in the later frames on-the-fly. Tracking-by-detection is one of the main research branches of online tracking. However, there still exist two issues in order to improve the performance. 1) The limited processing time requires the model to extract low-dimensional and discriminative features from the training samples. 2) The model is required to be able to balance both the prior and new objectives' appearance information in order to maintain the relocation ability and avoid the drifting problem. In this paper, we propose a real-time tracking algorithm called coupled randomness tracking (CRT) which focuses on dealing with these two issues. One randomness represents random projection, and the other randomness represents online random forests (ORFs). In CRT, the gray-scale feature is compressed by a sparse measurement matrix, and ORFs are used to train the sample sequence online. During the training procedure, we introduce a tree discarding strategy which helps the ORFs to adapt fast appearance changes caused by illumination, occlusion, etc. Our method can constantly adapt to the objective's latest appearance changes while keeping the prior appearance information. The experimental results show that our algorithm performs robustly with many publicly available benchmark videos and outperforms several state-of-the-art algorithms. Additionally, our algorithm can be easily utilized into a parallel program.

Horizon detection is useful in maritime image processing for various purposes, such as estimation of camera orientation, registration of consecutive frames, and restriction of the object search region. Existing horizon detection methods are based on edge extraction. For accuracy, they use multiple images, which are filtered with different filter sizes. However, this increases the processing time. In addition, these methods are not robust to blurting. Therefore, we developed a horizon detection method without extracting the candidates from the edge information by formulating the horizon detection problem as a global optimization problem. A horizon line in an image plane was represented by two parameters, which were optimized by an evolutionary algorithm (genetic algorithm). Thus, the local and global features of a horizon were concurrently utilized in the optimization process, which was accelerated by applying a coarse-to-fine strategy. As a result, we could detect the horizon line on high-resolution maritime images in about 50ms. The performance of the proposed method was tested on 49 videos of the Singapore marine dataset and the Buoy dataset, which contain over 16000 frames under different scenarios. Experimental results show that the proposed method can achieve higher accuracy than state-of-the-art methods.

In recent years, many variants of key point based image descriptors have been designed for the image matching, and they have achieved remarkable performances. However, to some images, local features appear to be inapplicable. Since theses images usually have many local changes around key points compared with a normal image, we define this special image category as the image with local changes (IL). An IL pair (ILP) refers to an image pair which contains a normal image and its IL. ILP usually loses local visual similarities between two images while still holding global visual similarity. When an IL is given as a query image, the purpose of this work is to match the corresponding ILP in a large scale image set. As a solution, we use a compressed HOG feature descriptor to extract global visual similarity. For the nearest neighbor search problem, we propose random projection indexed KD-tree forests (rKDFs) to match ILP efficiently instead of exhaustive linear search. rKDFs is built with large scale low-dimensional KD-trees. Each KD-tree is built in a random projection indexed subspace and contributes to the final result equally through a voting mechanism. We evaluated our method by a benchmark which contains 35,000 candidate images and 5,000 query images. The results show that our method is efficient for solving local-changes invariant image matching problems.

We address the problem of measuring matching similarity in terms of template matching. A novel method called two-side agreement learning (TAL) is proposed which learns the implicit correlation between two sets of multi-dimensional data points. TAL learns from a matching exemplar to construct a symmetric tree-structured model. Two points from source set and target set agree to form a two-side agreement (TA) pair if each point falls into the same leaf cluster of the model. In the training stage, unsupervised weak hyper-planes of each node are learned at first. After then, tree selection based on a cost function yields final model. In the test stage, points are propagated down to leaf nodes and TA pairs are observed to quantify the similarity. Using TAL can reduce the ambiguity in defining similarity which is hard to be objectively defined and lead to more convergent results. Experiments show the effectiveness against the state-of-the-art methods qualitatively and quantitatively.

In recent decade, many state-of-the-art algorithms on image classification as well as audio classification have achieved noticeable successes with the development of deep convolutional neural network (CNN). However, most of the works only exploit single type of training data. In this paper, we present a study on classifying bird species by exploiting the combination of both visual (images) and audio (sounds) data using CNN, which has been sparsely treated so far. Specifically, we propose CNN-based multimodal learning models in three types of fusion strategies (early, middle, late) to settle the issues of combining training data cross domains. The advantage of our proposed method lies on the fact that we can utilize CNN not only to extract features from image and audio data (spectrogram) but also to combine the features across modalities. In the experiment, we train and evaluate the network structure on a comprehensive CUB-200-2011 standard data set combing our originally collected audio data set with respect to the data species. We observe that a model which utilizes the combination of both data outperforms models trained with only an either type of data. We also show that transfer learning can significantly increase the classification performance.