Corners are an important kind of image feature and play a crucial role in solving various tasks. Over the past few decades, a great number of corner detectors have been proposed. However, there is no benchmark dataset with labeled ground-truth corners and unified metrics to evaluate their corner detection performance. In this paper, we build three benchmark datasets for corner detection. The first two consist of those binary and gray-value images that have been commonly used in previous corner detection studies. The third one contains a set of urban images, called the Urban-Corner dataset. For each test image in these three datasets, the ground-truth corners are manually labeled as objectively as possible with the assistance of a line segment detector. Then, a set of benchmark evaluation metrics is suggested, including five conventional ones: the precision, the recall, the arithmetic mean of precision and recall (APR), the F score, the localization error (Le), and a new one proposed in this work called the repeatability referenced to ground truth (RGT). Finally, a comprehensive evaluation of current state-of-the-art corner detectors is conducted.

/pdf/a-benchmark-for-the-evaluation-of-corner-detectors-2szkz7oe.pdf

A Benchmark for the Evaluation of Corner Detectors

The repeatability rate is an important measure for evaluating and comparing the performance of keypoint detectors. Several repeatability rate measurements were used in the literature to assess the effectiveness of keypoint detectors. While these repeatability rates are calculated for pairs of images, the general assumption is that the reference image is often known and unchanging compared to other images in the same dataset. So, these rates are asymmetrical as they require calculations in only one direction. In addition, the image domain in which these computations take place substantially affects their values. The presented scatter diagram plots illustrate how these directional repeatability rates vary in relation to the size of the neighboring region in each pair of images. Therefore, both directional repeatability rates for the same image pair must be included when comparing different keypoint detectors. This paper, firstly, examines several commonly utilized repeatability rate measures for keypoint detector evaluations. The researcher then suggests computing a two-fold repeatability rate to assess keypoint detector performance on similar scene images. Next, the symmetric mean repeatability rate metric is computed using the given two-fold repeatability rates. Finally, these measurements are validated using well-known keypoint detectors on different image groups with various geometric and photometric attributes. 

Two-Fold and Symmetric Repeatability Rates for Comparing Keypoint燚etectors

In this work, the Unity3D development platform and Web3D technology are integrated into the teaching method of advertising design to get rid of the issues due to lack of communication and efficacy through design of digital advertisement. Based on this, an approach for ingenious product design from nature is proposed, with an emphasis on attaining a functional interaction of aesthetic intent and geometric features and investigating the relationships amongst natural systems and designers in product design from nature (PDN). The ponderings and research findings for the methodologies associated with the proposed approach have been presented. This methodology is thought to significantly bring down the delivery time of ground-breaking design and development of products, both economically and technologically. The findings in comprehensive experiments demonstrates that interactive virtual technology can significantly enhance the efficacy and interaction of the whole system in the process of digital advertising design.

/pdf/research-on-the-implementation-of-advertising-design-1lgc0ml6.pdf

Research on the Implementation of Advertising Design Teaching Based on Unity3D Development Platform and Web3D Technology

A dedicated benchmark for contour-based corner detection evaluation

Corner is an import type of feature point of image and has been widely used in image analysis and vision tasks. To enhance the robustness to contour noise and geometrical transformations and meanwhile improve the localization accuracy, we develop a novel measure for contour-based corner finding algorithm by using the multi-scale tangent-to-point distance (MSTPD) technique. First, image contour extraction is conducted; second, the curvature of the extracted contour is computed with the tangent-to-point distance technique under different scales. By introducing relative distance, MSTPD is more robust to geometric transformations and by employing multi-scale technique MSTPD is also robust to contour noise and much accurate on localization. Experimental results show that MSTPD is a promising corner detection scheme compared with the other seven impressive corner detection methods based on two common evaluation criteria, that is, average repeatability and localization error. 

Feature detection using relative distance and multi-scale technique

In recent years, deep learning has made successful applications and remarkable achievements in the field of medical image registration, and the method of medical image registration based on deep learning has become the current research hotspot. However, the performance of convolutional neural networks may not be fully exploited due to neglect of spatial relationships between distant locations in the image and incomplete updates of network parameters. To avoid this phenomenon, MHNet, a multiscale hierarchical deformable registration network for 3D brain MR images, was proposed in this paper. This network was an unsupervised end-to-end convolutional neural network. After training, the dense displacement vector field can be predicted almost in real-time for the unseen input image pairs, which saves a lot of time compared with the traditional algorithms of independent iterative optimization for each pair of images. On the basis of the encoder-decoder structure, this network introduced the improved Inception module for multiscale feature extraction and expanding the receptive field and the hierarchical forecast structure to promote the update of the parameters of the middle layers, which achieved the best performance on the augmented public dataset compared with the existing four excellent registration methods.

An Unsupervised 3D Image Registration Network for Brain MRI Deformable Registration

Fine grain image is an important research field of image recognition, which can classify objects in more detail. Image feature extraction is usually implemented by bilinear network, which can meet the desired effect of feature extraction, but also discard some detailed feature information. In this study, a fine-grained classification optimization algorithm that combines the target detection method YOLOv7 and multi-feature fusion bilinear network is proposed to achieve fine-grained image classification. In the method, YOLOv7 is used to locate the target, and the recognition content is extracted according to the image detection frame; Then, the improved bilinear convolutional neural network structure is used to fuse the features of different channels and different levels of convolutional layers in the bilinear network, so as to obtain more feature information and improve the precision of fine-grained classification. Experimental results show that the classification results of this algorithm are better.

A fine-grained image classification method combining YOLOv7 and bilinear multi-level feature fusion

Person re-identification can quickly locate and find all the specified targets in complex scenes with multiple cameras, which has been widely applied in intelligent video surveillance and security system. As a state-of-art method proposed recently, ResNet exhibits promising performances on person re-identification. However, without intermediate fully connected layer, ResNet fails to fully grasp the global information in the detection process. To overcome the above problem, this paper proposes a person re-identification method named RG-BoTNet by fusing the Relation-aware Global Attention mechanism into BoTNet. Since relation-aware global attention is good at grasping the global information of the image, RG-BoTNet is powerful in extracting personal features. The good performances conducted on cuhk03 dataset in terms of Mean Average Precision (MAP) and Rank-1demonstrate the effectiveness of RG-BoTNet for person re-identification task.

A Person Re-identification Method Fusing Bottleneck Transformer andRelation-aware Global Attention

Feature detection is one of the basic research topics in the fields of image processing and corner that has been widely applied on vehicle detection, UAV image matching, camera calibration and so on. Particularly, a fast corner detector is of much benefit to many real-time tasks. A novel discrete curvature estimation method for corner detection based on the ratio of center distances of symmetric contour (RCDSC) is proposed in this paper. Benefiting from calculating the Euclidean distance twice only to estimate the discrete curvature at each point on a contour, RCDSC is much fast compared with other corner detectors. In addition, by choosing a relatively large radius of region of support and employing relative distance instead of absolute distance for constructing corner response function, RCDSC is also much robust to noises and local variations of contour. Extensive experiments exhibit the effectiveness and the efficiency of RCDSC in terms of average repeatability (AR), accuracy (ACU) and localization error (LE). 

Min Huang

Papers

Feature detection using relative distance and multi-scale technique

An Unsupervised 3D Image Registration Network for Brain MRI Deformable Registration

A fine-grained image classification method combining YOLOv7 and bilinear multi-level feature fusion

A Person Re-identification Method Fusing Bottleneck Transformer andRelation-aware Global Attention

An Efficient Corner Detector Using Ratio of Center Distance of Symmetric Contour Technique