Object representation based on local features is a topical subject in the domain of image understanding and computer vision. We discuss the defects of global features in present methods and the advantages of local features in object recognition, and briefly explore state-of-the-art recognition methods using local features, especially the main approaches of local feature extraction and object representation. To clearly explain these methods, the problem of local feature extraction is divided into feature region detection, feature region description, and feature space optimization. The main components and merits of these steps are presented. Technologies for object presentation are classified into three types: vector space, sliding window, and structure relationship models. Future development trends are discussed briefly.

Time series appear in many scientific fields and are an important type of data. The use of time series analysis techniques is an essential means of discovering the knowledge hidden in this type of data. In recent years, many scholars have achieved fruitful results in the study of time series. A statistical analysis of 120,000 literatures published between 2017 and 2021 reveals that the topical research about time series is mostly focused on their classification and prediction. Therefore, in this study, we focus on analyzing the technical development routes of time series classification and prediction algorithms. 87 literatures with high relevance and high citation are selected for analysis, aiming to provide a more comprehensive reference base for interested researchers. For time series classification, it is divided into supervised methods, semi-supervised methods, and early classification of time series, which are key extensions of time series classification tasks. For time series prediction, from classical statistical methods, to neural network methods, and then to fuzzy modeling and transfer learning methods, the performance and applications of these different methods are discussed. We hope this article can help aid the understanding of the current development status and discover possible future research directions, such as exploring interpretability of time series analysis and online learning modeling.

In modern medical diagnosis, the karyotype analysis for human chromosome is clinically significant for the diagnosis and treatment of genetic diseases. In such an analysis, it is critically important to segment the banded chromosomes. Chromosome segmentation, however, is technically challenging due to the variable chromosome features, the complex background noise, and the uneven image quality of the chromosome images. Owing to these technical challenges, the existing deep-learning-based algorithms would have severe overfitting problems and are ineffective in the segmentation task. In this paper, we propose a novel chromosome segmentation model with our enhanced chromosome processing, namely ChroSegNet. First, we develop enhanced chromosome processing techniques to realize the quality and quantity enhancement of the chromosome data, leading to the chromosome segmentation dataset for our subsequent network training. Second, we propose our novel chromosome segmentation model “ChroSegNet" based on U-Net. According to the characteristics of chromosome data, we have not only improved the baseline structure but also incorporate the hybrid attention module to ChroSegNet, which can extract the key feature information and location information of chromosome. Finally, we evaluated ChroSegNet on our chromosome segmentation dataset and obtained the MPA of 93.31% and the F1-score of 92.99%. Experimental results show that ChroSegNet not only outperforms the representative segmentation models in chromosome segmentation performance but also has a lightweight model structure. We believe that our proposed ChroSegNet is highly promising in future applications of genetic measurement and diagnosis.

Most matting researches resort to advanced semantics to achieve high-quality alpha mattes, and direct low-level features combination is usually explored to complement alpha details. However, we argue that appearance-agnostic integration can only provide biased foreground details and alpha mattes require different-level feature aggregation for better pixel-wise opacity perception. In this paper, we propose an end-to-end Hierarchical and Progressive Attention Matting Network (HAttMatting++), which can better predict the opacity of the foreground from single RGB images without additional input. Specifically, we utilize channel-wise attention to distill pyramidal features and employ spatial attention at different levels to filter appearance cues. This progressive attention mechanism can estimate alpha mattes from adaptive semantics and semantics-indicated boundaries. We also introduce a hybrid loss function fusing Structural SIMilarity (SSIM), Mean Square Error (MSE), Adversarial loss, and sentry supervision to guide the network to further improve the overall foreground structure. Besides, we construct a large-scale and challenging image matting dataset comprised of 59,600 training images and 1000 test images (a total of 646 distinct foreground alpha mattes), which can further improve the robustness of our hierarchical and progressive aggregation model. Extensive experiments demonstrate that the proposed HAttMatting++ can capture sophisticated foreground structures and achieve state-of-the-art performance with single RGB images as input.

This paper provides a full theoretical and experimental analysis of a serial algorithm for the point-in-polygon test, which requires less running time than previous algorithms and can handle all degenerate cases. The serial algorithm can quickly determine whether a point is inside or outside a polygon and accurately determine the contours of input polygon. It describes all degenerate cases and simultaneously provides a corresponding solution to each degenerate case to ensure the stability and reliability. This also creates the prerequisites and basis for our novel boolean operations algorithm that inherits all the benefits of the serial algorithm. Using geometric probability and straight-line equation $F\left(P\right)=(y_i-y_{i+1})(x_p-x_i)-(y_i-y_p)(x_{i+1}-x_i)$ , it optimizes our two algorithms that avoid the division operation and do not need to compute any intersection points. Our algorithms are applicable to any polygon that may be self-intersecting or with holes nested to any level of depth. They do not have to sort the vertices clockwise or counterclockwise beforehand. Consequently, they process all edges one by one in any order for input polygons. This allows a parallel implementation of each algorithm to be made very easily. We also prove several theorems guaranteeing the correctness of algorithms. To speed up the operations, we assign each vector a number code and derive two iterative formulas using differential calculus. However, the experimental results as well as the theoretical proof show that our serial algorithm for the point-in-polygon test is optimal and the time complexities of all algorithms are linear. Our methods can be extended to three-dimensional space, in particular, they can be applied to 3D printing to improve its performance.

Image forgeries can entirely change the semantic information of an image, and can be used for unscrupulous purposes. In this paper, we propose a novel image forgery localization network named as MUN, which consists of an M^3 encoder and a UN decoder. Firstly, the M^3 encoder is constructed based on a Multi-scale Max-pooling query module to extract Multi-clue forged features. Noiseprint++ is adopted to assist the RGB clue, and its deployment methodology is discussed. A Multi-scale Max-pooling Query (MMQ) module is proposed to integrate RGB and noise features. Secondly, a novel UN decoder is proposed to extract hierarchical features from both top-down and bottom-up directions, reconstructing both high-level and low-level features at the same time. Thirdly, we formulate an IoU-recalibrated Dynamic Cross-Entropy (IoUDCE) loss to dynamically adjust the weights on forged regions according to IoU which can adaptively balance the influence of authentic and forged regions. Last but not least, we propose a data augmentation method, i.e., Deviation Noise Augmentation (DNA), which acquires accessible prior knowledge of RGB distribution to improve the generalization ability. Extensive experiments on publicly available datasets show that MUN outperforms the state-of-the-art works.