Aiming to improve quality indicator prediction for the tobacco drying process in a cigarette factory, this study proposes an adaptive-weighted multi-level feature evidence and redundancy-penalty framework (F-XMF). To alleviate the limitations of isolated feature evaluation in complex industrial data, F-XMF jointly considers the direct contribution of individual features, the contribution of feature interactions, and the redundant dependency among candidate features. First, eXtreme Gradient Boosting (XGBoost) is used to estimate the importance of individual features and obtain an initial main-effect feature subset. Meanwhile, an Attentional Factorization Machine (AFM) is employed to identify high-contribution second-order interactions and derive interaction-related evidence. On this basis, conditional mutual information (CMI) is introduced to characterize redundant dependency among candidate features, and a Differential Redundancy Penalty Mechanism is further developed to impose differentiated penalties on highly overlapping features while preserving informative variables with meaningful interaction utility. Finally, a validation-based adaptive weight fusion strategy is used to integrate the main-effect subset and the interaction-related subset, yielding a compact and informative final feature subset. Experimental results on real tobacco-processing data show that F-XMF consistently reduces Mean Absolute Error (MAE) in ablation comparisons and achieves superior performance against multiple representative baseline methods. Mechanism validation further demonstrates that the proposed framework can effectively refine the candidate subset through interaction enhancement and redundancy suppression, showing good interpretability and practical value for outlet moisture content prediction in the tobacco drying process.

This paper presents a novel optimized quantization constraint set, acting as an add-on to existing DCT-based image restoration algorithms. The constraint set is created based on generalized Gaussian distribution which is more accurate than the commonly used uniform, Gaussian or Laplacian distributions when modeling DCT coefficients. More importantly, the proposed constraint set is optimized for individual input images and thus it is able to enhance image quality significantly in terms of signal-to-noise ratio. Experimental results indicate that the signal-to-noise ratio is improved by at least 6.78% on top of the existing state-of-the-art methods, with a corresponding expense of only 0.38% in processing time. The proposed algorithm has also been implemented in GPU, and the processing speed increases further by 20 times over that of CPU implementation. This makes the algorithm well suited for fast image retrieval in security and quality monitoring system.

Homography has wide applications in aerial photographic surveys, camera calibration, traffic scene analysis, and sports science, such as player and team performance evaluation. Unlike the mainstream homography that utilizes points as matching features, homography estimation for sports and traffic video can achieve higher accuracy and speed by utilizing straight lines in the scenes, which convey more information than points. Owing to the more stringent requirement of accuracy and computational speed for advanced video analysis, this paper presents a novel homography computational algorithm. Three major novelties are proposed and validated, which are multiple points Hough transform for straight line extraction, correspondence initialization by angle to estimate a set of quasi-optimal solutions, and the feature correspondences optimization to achieve a minimized error using genetic algorithm. With these contributions, the experiments have shown that the proposed algorithm can improve the homography computational accuracy by up to 130% and reduce the processing time by up to 96% over the state-of-the-art algorithms for the same purposes.

Construction and sensing within malleable interfaces is usually limited by a number of constraints. Building the interface from diverse combinations of conductive and nonconductive soft materials, such as fabrics or foams combined with various sensors, complicates the manufacturing process and offers limited options in shaping. In this paper we propose "Unique Topologies for Acoustic Propagation" (UTAP), a novel approach for algorithmic design of malleable tangible interfaces. A fundamental feature of our approach is the implementation of algorithmically generated topologically distinct lattices that, attached to piezoelectric (PZT) transducers, allow us to sense and recognize changes in a modulated acoustic signal on deformation and classify it into different interaction states. Our systematic approach to manufacturing malleable interfaces opens possibilities to design shapes that allow implementation in a wide range of potential use cases. We demonstrate the UTAP approach on multiple interfaces assembled using laser cut and 3D printed lattices in conjunction with silicon compound moulding. Finally, we present a technical evaluation of our method based on studies of four distinct interface designs, assessing performance in sensing and localising simple deformations such as pressing on single and multiple spots, as well as different force levels and actions, including bending and twisting.

Bleed-through effect is one of the most common degradations in old documents, even in today's newspapers. This effect must be removed for hancing human and automatic readability. The two images scanned from the recto and verso pages of a document can be treated as a linear combination of clean text images from two sides. In this paper, we introduce a nonnegative least-correlation approach to demix bleed-through text images. Experiments have been conducted on both synthetic and real world images. In synthetic case, our method can recover the source text images exactly. Under real world conditions, our approach also performs well. In addition, our approach is computationally efficient and does not need any postprocessing task.