A color design process for fabric images can resort to a solution of transfer problem based on given themes. Usually, the contains an image segmentation phase and construction phase. In this paper, novel method is proposed. Compared with classical methods, new has following three main innovations. First, method, in its phase, follows assumption that be decomposed into cartoon texture components, which means segmentation, incorporates decomposition process. The advantage innovation component...

Training models with longer in-context lengths is a significant challenge for multimodal model due to substantial GPU memory and computational costs. This exploratory study does not present state-of-the-art models; rather, it introduces an innovative method designed increase text length in multi-modality large language (MLLMs) efficiently. We Visualized In-Context Text Processing (VisInContext), which processes long using visual tokens. technique significantly reduces usage floating point...

In order to improve the precision of visible and infrared (VIS/IR) image registration, an registration method based on visual salient (VS) features is presented. First, a VS feature detector modified attention model presented extract points. Because iterative, within-feature competition used in models time consuming, alternative fast (FVS) proposed make more efficient. Then, descriptor-rearranging (DR) strategy adopted describe This combines information both IR its negative overcome contrast...

Fabric image recolorization is widely used in assisting designers to generate new color design proposals for fabric. In this paper, a method proposed. Different from classical methods, which need some complicated interactive operations users, our proposed can achieve automatic of images. The contains three sequential phases: phase extracting representative colors fabric images; an segmentation phase; and reconstruction by using given themes. Integrated with intrinsic decomposition, model...

Discusses the text segmentation of Manchu character recognition process, improved traditional projection method, improves accuracy Manchu. The nearest neighbor and support vector machine, Bayes three combined with characteristics Manchu, a design based on printed system multi classifier fusion decision, realize image preprocessing, other functions. experimental results show that, has achieved good effect

This paper studied the problem of detecting and extracting insulator from aerial image proposed a new robust detection method based on visual attention mechanism. Firstly, local global saliency for images were calculated. Secondly, maps combined together to get final map. Thirdly, region was extracted map non-insulator excluded by pro-processing method. Lastly, converted into binary separated background adding original image. Experimental results showed that could extract precisely, it...

Quantitative ultrasound (QUS), which attempts to extract quantitative features from the US radiofrequency (RF) or envelope data for tissue characterization, is becoming a promising technique noninvasive assessments of liver fibrosis. However, number feature variables examined and finally used in existing QUS methods typically small, limiting diagnostic performance. Therefore, this paper devises new multiparametric (MP-QUS) method enables extraction large RF signals allows use...

Aging population and empty-nesters are two big challenges in modern healthcare. Fall incidents can cause various physical injuries serious consequence without receiving timely assistance. Therefore, fall detection movement classification have very high research value application significance. This paper aims to study the optimum feature subset of falls put forward an improved approach detect falls. A set twelve motion features eleven kinds activities extracted from different parts body....

Manchu character recognition is important in protecting and researching culture history. Previous methods of are mainly based on conventional machine learning using shallow artificial selection features, thus results unsatisfactory. The method with convolutional neural networks achieves high accuracy optical as the convolution operators can automatically extract deep structure features. network needs input images fixed size, but a kind phonemic language, word has an arbitrary length. So it...

In this paper we present a novel combined tracking algorithm based on moving object detection technology. Particle filtering can work well when the gets an occlusion, it has difficulty in satisfying requirement of real-time computing. Meanshift solve problem easily, poor rubustness during mutual occlusion. meantime, because backgrounds many scenes include complex objects, methods only using color, texture and shape feature often have rubustness. Aiming at all above problems, detect codebook...

Precise Manchu character segmentation of segmentation-based recognition methods is difficult to realize because complex language spelling rules and existence multi fonts. To avoid the influence incorrect segmentation, this work proposes idea segmentation-free recognize word instead characters. In addition, an end-to-end 9-layer convolutional neural network proposed automatically extract deep hierarchy features on image. The model applied words with 12 fonts evaluate its ability multi-font...

The convex nonnegative matrix factorization (CNMF) is a variation of (NMF) in which each cluster expressed by linear combination the data points and point represented centers. When there exists nonlinearity manifold structure, both NMF CNMF are incapable characterizing geometric structure data. This paper introduces neighborhood preserving (NPCNMF), imposes an additional constraint on that can be as its neighbors. Thus our method able to reap benefits purpose structure. An efficient...

Aging population is a big challenge in modern healthcare. Injuries caused by fall incidents are great threats to the elderly. As consequence, detection and movement classification have very high research value application significance. This paper aims study optimum feature subset of falls verify effectiveness proposed features through experiments. A set twelve eleven kinds activities extracted from four sensor data firstly. Then, an improved classifier based on Bayesian framework for...

Deformable registration (DR) is a key technology in the medical field. However, many of existing DR methods are time-consuming and accuracy needs to be improved, which prevents their clinical applications. In this study, we propose parallel block matching algorithm for lung CT image registration, sum squared difference metric modified as cost function moving least squares approach used generate full displacement The implemented on Graphic Processing Unit (GPU) with Compute Unified Device...

There is always much difficult in the MR image segmentation. Although fuzzy c-means(FCM) clustering algorithm has been widely used field of segmentation study, some inherent deficiencies this especially high cost computation made to be practice. A novel algorithm, based on kernel c-means (KFCM) and k-nearest neighbor (KNN) proposed improve performance In statistical gray level histogram KFCM speed up algorithm. Furthermore, spatial information also considered by methods. With methods each...