The Coons patch is a classical method for constructing smooth two-dimensional manifold mappings through boundary curve interpolation.It is widely applied in fields such as computer graphics,product shape de-sign,and isogeometric analysis.However,due to factors such as complex boundary curve shapes and significant curvature variations,the Jacobian determinant of a Coons mapping may degenerate or even change sign in certain regions,causing a loss of regularity.This can severely affect the quality of the surface and compromise the nu-merical stability of subsequent analyses.Developing efficient and reliable methods for regularity verification is of great importance for achieving high-quality parameterizations.To address the regularity verification problem of spatial Coons patches,a series of theoretical conditions and algorithms is proposed.First,based on the condition of the rank of Jacobian matrix,a sufficient criterion for ensuring the regularity of Coons mappings is derived.Second,for Coons patches expressed in Bézier form,a regularity verification method based on the Bézier coeffi-cients of the Jacobian determinant is proposed,transforming the verification task into a set of coefficient sign-consistency constraints,which significantly improves computational efficiency and geometric interpretabil-ity.Finally,a subdivision strategy for Bézier surfaces is introduced to construct a class of necessary and sufficient conditions,enabling the global verification problem to be reduced to local subdomain checks.The proposed Jaco-bian coefficient extraction framework can be extended to multi-patch B-spline surfaces with general topological structures,enhancing the generality and adaptability of the proposed method.Numerical experiment results show that for multi-patch B-spline surfaces consisting of 1 752 and 6 800 bi-cubic Bézier elements,the proposed algo-rithm completes the regularity verification within 0.112 seconds and 0.402 seconds,respectively,meeting the performance requirements of real-time engineering applications.
With the popularization of virtual reality technology,panoramic image quality assessment faces new challenges.Addressing the issue that existing full-reference methods are mostly simple extensions of traditional indicators,while no-reference methods,although relying on deep learning,generally lack interpretability,this study proposes a full-reference panoramic image quality assessment method based on visual perception mechan-isms.First,it integrates the advantages of equidistant rectangular projection and cube mapping projection,and calculates the structural similarity index in the discrete cosine transform domain to more accurately capture key structural features.Then,it introduces the equatorial bias factor to enhance the method's conformity to human visual attention characteristics.Experiments on the CVIQD and MVAQD databases,and comparisons with exist-ing panoramic image quality assessment algorithms,show that the proposed method outperforms the comparative methods in both Pearson linear correlation coefficient(PLCC)and Spearman rank correlation coefficient(SRCC).This method can effectively improve the accuracy and perceptual consistency of panoramic image quality as-sessment,providing technical support for optimizing the virtual reality user experience.
To address the issues of feature information loss,inaccurate threshold settings,and detail informa-tion loss due to low resolution during the fusion process of traditional Dynamic Threshold Neural P(DTNP)systems,a multi-modal medical image super-resolution fusion method based on a feature link model opti-mization,called Feature Link DTNP(FL-DTNP)system,is proposed.This method uses adaptive link strength for image feature extraction.First,bicubic interpolation and Non-Subsampled Shearlet Transform(NSST)are applied to enhance and decompose the image.Then,the FL-DTNP system is used to fuse the detail and edge information of the high-frequency sub-bands.For the low-frequency sub-bands,a visually significant weighted local energy approach—based on an eight-neighborhood Laplacian weighting—is ap-plied to merge the low-frequency sub-bands.Finally,inverse NSST is used to reconstruct the high-frequency and low-frequency sub-bands.Experiments were conducted on the Harvard dataset,comparing the proposed method with eight classic fusion methods based on multi-scale transformations,deep learning,and sparse representation across seven evaluation metrics.The results demonstrate that the proposed method exhibits outstanding performance and is capable of generating high quality fused images.
Blind Image Quality Assessment(BIQA)aims to simulate human prediction of image quality distor-tion levels and provide quality scores.However,existing unimodal-based BIQAs have limited representational ability when facing complex contents and distortion types,and the predicted scores also fail to provide explana-tory descriptions which further affects the credibility of their prediction results.To address these challenges,we propose an eXplainable Blind Image Quality Assessment(xBIQA)guided by Large Language Model(LLM).Our method leverages image distortion and overall description to generate global quality text,while local quality text is produced to provide detailed descriptions of specific areas.These global texts,local texts,and prompts are then jointly fed into an LLM to generate detailed semantic features.Compared to traditional BIQA methods based on a single image modality,our approach demonstrates that LLMs can effectively produce text descriptions highly correlated with image quality,thereby enhancing the performance of BIQA models based on multimodal learning.Then,we align and fuse the text semantic features and the image texture features,and regress to obtain the image quality score,while outputting its corresponding quality explanatory description.Experimental results show that our xBIQA performs best on the KonIQ-10k and LIVE Challenge datasets,with improvements of 1.64%and 2.60%in the SRCC metric,respectively.
Existing visualization methods for cloud service networks struggle to collaboratively present dual hierarchical structures of spatial and logical deployments along with diverse device attributes,which ad-versely affects operational efficiency.This paper proposes a network simplification algorithm that utilizes node aggregations on equivalent node units to preserve network characteristics meanwhile reducing network scale.Then,a collaborative layout algorithm that integrates weighted Voronoi-based nested segmentation with a strong-boundary-constrained force-directed layout is designed to achieve the seamless presentation of dual hierarchical structures.Finally,a set of visual encodings and interactions are designed to depict device attributes.Experiments and case studies were conducted on six real-world cloud service networks of varying scales.The results demonstrate that the proposed method achieves superior performance across multiple as-pects:in terms of algorithmic performance,the alarm aggregation rate remained at 0%,the average degree deviation ranged from 6.43%to 13.90%,and the node simplified rate reached 89.38 to 94.93;in terms of visual perception efficiency,the visual center index ranged from 0.86 to 0.91,and the node overlap rate was controlled between 0%to 0.37%;regarding practical operational support,volunteers completed four tasks derived from reference real-world alarm analysis scenarios in an average time of 5.96 s to 8.12 s,achieving an accuracy of 95.00%to 100.00%,and the average scores of three Likert-scale evaluations of visualization effectiveness ranged from 4.1 to 4.4.All results outperform the comparative methods.
