A lightweight 3D anomaly detection method with rotationally invariant features

Great progress has been made in learning-based object detection methods in the last decade. Two-stage detectors often have higher detection accuracy than one-stage detectors, due to the use of region of interest (RoI) feature extractors which extract transformation-invariant RoI features for different RoI proposals, making refinement of bounding boxes and prediction of object categories more robust and accurate. However, previous RoI feature extractors can only extract invariant features under limited transformations. In this paper, we propose a novel RoI feature extractor, termed Semantic RoI Align (SRA), which is capable of extracting invariant RoI features under a variety of transformations for two-stage detectors. Specifically, we propose a semantic attention module to adaptively determine different sampling areas by leveraging the global and local semantic relationship within the RoI. We also propose a Dynamic Feature Sampler which dynamically samples features based on the RoI aspect ratio to enhance the efficiency of SRA, and a new position embedding, i.e., Area Embedding, to provide more accurate position information for SRA through an improved sampling area representation. Experiments show that our model significantly outperforms baseline models with slight computational overhead. In addition, it shows excellent generalization ability and can be used to improve performance with various state-of-the-art backbones and detection methods. The code is available at https://github.com/cxjyxxme/SemanticRoIAlign.

In this paper we present an effective method for object detection in digital images. Our approach is fast, stable and easy to implement. It is motivated by a strong physical and mathematical basis, which ensures an invariance of the recognition with respect to illumination, rotations, scaling and translations. In addition, there are no assumptions on the geometry of the object which has to be recognized. Our method extracts distinctive points of the image and approximates a small pixel neighborhood by polynomials. The corresponding polynomial coefficients are used to compute invariant feature vectors for solving point correspondences in order to calculate an optimal prototype fitting.

The malicious code on the network is increasingly rampant that the traditional detection method of characteristic code has been difficult to deal with malicious code, with features of various variants, deformations and other problems. In this paper we present a new static analysis model based on software fingerprint to distinguish malicious codes. Through obtaining the software call graph by disassembling the binary file and mapping it as an image, shape moments can be obtained as the software fingerprint based on the retrieval theory of content image, combined with moment theory and the image's color, texture and shape features. The idea of pattern matching is used to measure the extracted software fingerprint similarity to determine whether it is malicious code or not. Then, we analyze the collected program samples. Test and verify whether the program has good performance in uniqueness, invariability and sensibility. With the rapid popularization of Internet and corporate information technology acceleration, the computer is a great convenience to people's lives, whether it is shopping, leisure or work and more obvious the importance of the Internet, but due to the openness of the Internet and flexible application and operating system vulnerability so that people can enjoy the benefits brought by the computer at the same time, also is experiencing distress and abuse of all kinds of malicious code threats to network security events increased year by year. Network security incidents, the most serious harm caused by malicious code, causing huge economic losses to the country as a whole, society and the individual, information security has become a major challenge facing. Domestic and foreign researchers turned to the semantics of malicious code, trying to judge the signatures of the two deformation malicious code through the instruction-level semantics rather than program syntax, further can determine whether the deformation of the malicious code. Trying to evade detection of malicious code, malicious codes are disassembled and for standardization or stack analysis system call for the use of fuzzy transformation technology to discriminate. In recent years, the researchers also uses malicious code detection technology engineering methodology, based on feature detection based on the application of malicious code detection technology based on data mining and machine learning. However, while foreign research scholars during the malicious code detection and anti-virus software R & D, the attacker malicious code using anti-debugging techniques, anti-Hook technology, to detect whether the code being debugged, the to find themselves debugger or analysis environment, malicious code using fuzzy transform technique and a series of anti-debugging measures show some non-anomalous behavior of the code, so as to protect their own purposes. In this process, we need manually assistive technology. Most automated virus analysis software only to capture some of the behavior, security experts need further analysis and screening of experimental results, and finally determine the extent of the harm of malicious behavior. In summary, the malicious code on the network more and more rampant code detection method for the traditional characteristics of difficult to deal with malicious code variants, deformation problem, solve the problem of the detection of unstable deformation caused due to malicious code upgrade has become a research focus and difficulty, but also of the issues that must be resolved key issues.

This article introduces the application of a physics-based symbolic image partitioning method to detect targets in synthetic aperture radar (SAR) imagery. ‘Targets’ in this case refer to vehicular objects which produce a distinct radar return pattern, and have spatial characteristics that are known a priori. The proposed Rotationally Invariant Symbolic Histogram (RISH) detection method co-analyses both target and speckle statistics, and significantly reduces computational requirements by partitioning the data into a discrete number of state representations. RISH requires only one pass for robust detection, unlike other SAR detection methods which rely on difference metrics calculated using multiple passes. To improve performance in high-resolution data, RISH uses a weighted feature extraction algorithm to avoid the common requirement of processing each pixel of the image equally. The weighted structure extracts geometrically undefined and rotationally invariant target features. This article details the analysis of 24 experimentally obtained very high-frequency (VHF)-band SAR magnitude images using this novel approach to SAR target detection. In localizing small (~8.4 m2) foliage-concealed targets, without the aid of pre-processing, this method results in high performance characteristics (90% true positive) with a low Type-II error rate of 6.4 false alarms per 1 × 106 m2. With the addition of change detection, RISH lowers the error rate by 85%.

In this work, fast and efficient appearance-based methods for visual loop-closure detection are proposed. The widely used technique based on the Bag-of-Words image representation has shown some limitations especially with aliasing problem. In this work, however, an appearance-based approach for loop closure detection using local invariant and colour features is proposed. The first technique uses Bayes Decision Theory for loop closure detection based on Gaussian Mixture Model (GMM). A new technique based on the combination of GMM with the KD-Tree data structure is presented as well. The techniques have been validated using monocular image sequences from several environments.

While current state-of-the-art Remote Sensing Change Detection (RSCD) methods can achieve impressive results on individual datasets, they become unreliable in unseen environments and imaging conditions, with performance metrics declining by as much as 60% to 80%. Simultaneously, variable environments and complex imaging conditions are the main characteristics of remote sensing data, calling for generalizable RSCD methods. To address this issue, we propose a novel RSCD method capable of domain generalization—CDDGNet. This method is based on causal decoupling theory, which progressively decouples invariant change features from variable domain features to extract generalizable characteristics. This enables a network trained on a single domain to accurately identify change regions in other domains. Specifically, firstly, the Causal Feature Adaptation Module is proposed to preliminarily decouple and simplify feature information during the encoding process by using wavelet transformation and feature energy spectralization methods. Secondly, the Causal Feature Fusion Module is presented to fully decouple features and aggregate significant change features during the decoding process through frequency domain processing and feature re-attention mechanisms. Thirdly, the Decoupling Effect Loss Function is proposed to optimize the process by evaluating the effectiveness of causal decoupling. Extensive experiments have shown that our model significantly outperforms existing methods across multiple groups of generalization tasks with varying levels of difficulty.

In recent years, Ambient Intelligence (AmI) has attracted increased attention within the advanced technology industry in an effort to modernize and develop a more intelligent and reliable information system. Technologies to detect a specific object in images are expected to further expand to wide range of applications, such as car detection functions for Intelligent Transport System (ITS) and other systems. Computer vision and pattern recognition are emerging fast and will continue to grow together with local feature detection methods. In this paper, we propose an object detection and tracking system which is based on Scale Invariant Feature Transform (SIFT) and Speed Up Robust Features (SURF) feature extraction methods. From the evaluation results, we observe that the accuracy of matched keypoints of SURF algorithm are higher than SIFT.

In this paper, an affine invariance feature detection method based on Scale Invariant Feature Transform (SIFT) and Maximally Stable Extremal Regions (MSER) is proposed. Classical SIFT algorithm is not robust to affine deformations, because it is based on DOG detector which extracts circle regions for keypoint location. In order to overcome this disadvantage, DOG detector in conventional SIFT algorithm is replaced by MSER detector which is robust to affine deformation. Then these regions are normalized and extracted using SIFT. Simulation studies are carried out to show the effectiveness of the proposed method to affine transform in comparison to traditional SIFT algorithm.

Diffusion-based generative models have significantly advanced the field of image synthesis, presenting additional challenges regarding the integrity and authenticity of digital images. Consequently, the identification of AI-generated images has become a critical problem in image forensics. However, there is a lack of literature addressing the detection of images generated by diffusion models. In this paper, our focus is on developing a model capable of detecting images generated through both GAN techniques and diffusion models. We propose DiffCoR, a novel detection method for identifying AI-generated images. It consists of two main modules: Stable Diffusion Processing (SDP) and Image Representation Learning (IRL). The SDP module uses a pre-trained Stable Diffusion model to reconstruct input images via reverse diffusion and captures subtle manipulations through reconstruction discrepancies. The IRL module applies self-supervised learning with Latent Consistency Loss (LCL) to extract robust, invariant features, ensuring consistent latent representations across augmented views. We also incorporate frequency domain analysis using Discrete Fourier Transform (DFT) to enhance manipulation detection. Additionally, we introduce ForensicsImage, a publicly available dataset of over 400,000 real and AI-generated images from LSUN-Bedroom, CelebA-HQ, CelebDFv2, and various diffusion models. Experiments on ForensicsImage and GenImage show that DiffCoR achieves state-of-the-art performance, with strong cross-dataset generalization, making it suitable for real-world use in digital forensics, content verification, and social media moderation. Our code and dataset are available at https://github.com/nhantran214/diffcor.

Multi-camera 3D object detection aims to detect and localize objects in 3D space using multiple cameras, which has attracted more attention due to its cost-effectiveness trade-off. However, these methods often struggle with the lack of accurate depth estimation caused by the natural weakness of the camera in ranging. Recently, multi-modal fusion and knowledge distillation methods for 3D object detection have been proposed to solve this problem, which are time-consuming during the training phase and not friendly to memory cost. In light of this, we propose PromptDet, a lightweight yet effective 3D object detection framework motivated by the success of prompt learning in 2D foundation model. Our proposed framework, PromptDet, comprises two integral components: a general camera-based detection module, exemplified by models like BEVDet and BEVDepth, and a LiDAR-assisted prompter. The LiDAR-assisted prompter leverages the LiDAR points as a complementary signal, enriched with a minimal set of additional trainable parameters. Notably, our framework is flexible due to our prompt-like design, which can not only be used as a lightweight multi-modal fusion method but also as a camera-only method for 3D object detection during the inference phase. Extensive experiments on nuScenes validate the effectiveness of the proposed PromptDet. As a multi-modal detector, PromptDet improves the mAP and NDS by at most 22.8% and 21.1% with fewer than 2% extra parameters compared with the camera-only baseline. Without LiDAR points, PromptDet still achieves an improvement of at most 2.4% mAP and 4.0% NDS with almost no impact on camera detection inference time. We will release our code.

Purpose – To build a persistent map with visual landmarks is one of the most important steps for implementing the visual simultaneous localization and mapping (SLAM). The corner detector is a common method utilized to detect visual landmarks for constructing a map of the environment. However, due to the scale-variant characteristic of corner detection, extensive computational cost is needed to recover the scale and orientation of corner features in SLAM tasks. The purpose of this paper is to build the map using a local invariant feature detector, namely speeded-up robust features (SURF), to detect scale- and orientation-invariant features as well as provide a robust representation of visual landmarks for SLAM. Design/methodology/approach – SURF are scale- and orientation-invariant features which have higher repeatability than that obtained by other detection methods. Furthermore, SURF algorithms have better processing speed than other scale-invariant detection method. The procedures of detection, description and matching of regular SURF algorithms are modified in this paper in order to provide a robust representation of visual landmarks in SLAM. The sparse representation is also used to describe the environmental map and to reduce the computational complexity in state estimation using extended Kalman filter (EKF). Furthermore, the effective procedures of data association and map management for SURF features in SLAM are also designed to improve the accuracy of robot state estimation. Findings – Experimental works were carried out on an actual system with binocular vision sensors to prove the feasibility and effectiveness of the proposed algorithms. EKF SLAM with the modified SURF algorithms was applied in the experiments including the evaluation of accurate state estimation as well as the implementation of large-area SLAM. The performance of the modified SURF algorithms was compared with those obtained by regular SURF algorithms. The results show that the SURF with less-dimensional descriptors is the most suitable representation of visual landmarks. Meanwhile, the integrated system is successfully validated to fulfill the capabilities of visual SLAM system. Originality/value – The contribution of this paper is the novel approach to overcome the problem of recovering the scale and orientation of visual landmarks in SLAM tasks. This research also extends the usability of local invariant feature detectors in SLAM tasks by utilizing its robust representation of visual landmarks. Furthermore, data association and map management designed for SURF-based mapping in this paper also give another perspective for improving the robustness of SLAM systems.

At present, the detection of surface defects of paper-plastic composite bags still mainly depends on manual visual inspection, which is very inefficient and easy to cause false detection. Based on the algorithm which combines edge detection and adaptive region growth, binary images of the surface defect areas is extracted, six types of shape features and seven types of invariant moment features are defined by calculation as the basis to complete the effective extraction of the surface features, then a paper-plastic composite bags surface defect detection platform is built to complete the defect detection experiment, the defect screening experiment was completed by the rapid screening method of its surface defect image based on gradient projection difference. At the same time, based on Genetic Algorithm (GA), Particle Swarm Optimization, and Grey Wolf Optimization, parameter optimization models of Support Vector Machine were established, respectively. Through experimental comparison, some results show that the defect detection and classification accuracy rate reach 96.83% based on the GA, and the detection speed with the screening method is 3.75 times faster than that without the screening method. Therefore, the reliability of the image classification and screening method for bags surface defects proposed in this study is verified.

We propose a new efficient technique for localization of faces in arbitrary images. The technique is based on segmentation of images into skin colored blobs, which is followed by computation of scale, translation and rotation invariant moment-based features to learn a statistical model of faces and non-face regions. The superiority of the method to the state-of-the-art face detection methods is its ability to detect non-frontal faces in a person independent way. Experimental results on CVL database show that the proposed algorithm gives higher true positive rates as compared to the well-known Viola-Jones face detector.

Pavement cracks are an important hidden danger for driving safety. Rapid and accurate detection of road crack types and effective treatment can greatly reduce the incidence of traffic accidents, which has important practical significance. Based on the advantages of scale invariant feature transform (SIFT) and back-propagation neural network (BP) algorithm, this paper proposes a method of road crack detection and recognition for four types of cracks, such as cracks, longitudinal cracks, transverse cracks and massive cracks.Firstly, the crack image is denoised and enhanced; Then scale invariant feature transform method is used to extract the feature point information of the route crack image; Finally, Back propagation neural network is used for training and recognition. The results show that this method can quickly and accurately identify the crack type, in which the accuracy of training set is 98.79%, and the accuracy of recognition set is 86.47%. The results show that the combination of scale invariant feature transform and back propagation neural network algorithm can effectively improve the efficiency of pavement crack type detection.

We present a theory and a method for simultaneous detection of local spatial and temporal scales in video data. The underlying idea is that if we process video data by spatio-temporal receptive fields at multiple spatial and temporal scales, we would like to generate hypotheses about the spatial extent and the temporal duration of the underlying spatio-temporal image structures that gave rise to the feature responses. For two types of spatio-temporal scale-space representations, (i) a non-causal Gaussian spatio-temporal scale space for offline analysis of pre-recorded video sequences and (ii) a time-causal and time-recursive spatio-temporal scale space for online analysis of real-time video streams, we express sufficient conditions for spatio-temporal feature detectors in terms of spatio-temporal receptive fields to deliver scale covariant and scale invariant feature responses. A theoretical analysis is given of the scale selection properties of six types of spatio-temporal interest point detectors, showing that five of them allow for provable scale covariance and scale invariance. Then, we describe a time-causal and time-recursive algorithm for detecting sparse spatio-temporal interest points from video streams and show that it leads to intuitively reasonable results.