Search

General learning algorithms trained on a specific dataset often have difficulty generalizing effectively across different domains. In traditional pattern recognition, a classifier is typically trained on one dataset and then tested on another, assuming both datasets follow the same distribution. This assumption poses difficulty for the solution to be applied in real-world scenarios. The challenge of making a robust generalization from data originated from diverse sources is called the domain adaptation problem. Many studies have suggested solutions for mapping samples from two domains into a shared feature space and aligning their distributions. To achieve distribution alignment, minimizing the maximum mean discrepancy (MMD) between the feature distributions of the two domains has been proven effective. However, this alignment of features between two domains ignores the essential class-wise alignment, which is crucial for adaptation. To address the issue, this study introduced a discriminative, class-wise deep kernel-based MMD technique for unsupervised domain adaptation. Experimental findings demonstrated that the proposed approach not only aligns the data distribution of each class in both source and target domains, but it also enhances the adaptation outcomes. [ABSTRACT FROM AUTHOR]

This paper proposes an efficient method for rectifying distorted document images via deep learning, ultimately improving the legibility of graphics and text in documents. The framework comprises two interconnected UNets, working in tandem to predict a 3D coordinate map and a forward map for the input distorted document image, respectively. At the beginning of the process, a page mask is predicted and used as input to both U-Nets to help improve the performance of their tasks. In the last step, the predicted forward map is transformed into a corresponding backward map, which is utilized to rectify the distorted image. The experimental results not only reveal that the predicted page masks and 3D coordinate maps significantly enhance the accuracy of predicting forward maps for subsequent rectification but also demonstrate satisfactory results both globally and locally. [ABSTRACT FROM AUTHOR]

Unsupervised domain adaptation (UDA) mainly explores how to learn domain-invariant features from the source domain when the target domain label is unknown. To learn domain-invariant features requires aligning the distribution of samples from two domains in the feature space, which can be achieved by minimizing the maximum mean discrepancy (MMD) of samples from the two domains. However, there is still no effective way to find the best parameter values of MMD. Such a problem is addressed in the MMD with deep kernels (MMD-D), whose optimal parameters can be obtained through training. This study proposes a method of domain-invariant feature learning for UDA, whose architecture, named MMDDCDA, comprises an MMD-D module and a cross domain adaptation (CDA) module. MMDDCDA performs alternating training similar to adversarial training to alternately boost the power of the two modules. To our knowledge, this is the first UDA method that performs such alternating training on a UDA architecture using MMD with deep kernels. Experimental validation showed that the proposed method yields state-of-the-art results among UDA methods using other MMD variants and some UDA benchmarks. [ABSTRACT FROM AUTHOR]

This paper proposes a flexible example-based color transfer system, providing an automatic mode and an advanced mode, for novices and expert users, respectively. The experimental results show that the proposed color transfer system not only can introduce natural results with simple operation by novices in the first use, but also can produce various desired results by users with short-term learning. [ABSTRACT FROM AUTHOR]

Due to declining trading volume growth in e-commerce platforms, physical channels have attracted considerable investments from various large international companies (e.g. Alibaba, JD, Walmart, Wanda, and Wuzhou International). However, e-commerce platforms can track consumers' behaviors (attraction to landing page design, clicks on certain products, consumer behavior trajectory tracking, clicks on advertisements, and internal link optimization of product pages), a feat unachievable in current physical channels. Consequently, this study attempted to apply the characteristics of online channels in a physical channel by using image object tracking and image detection techniques. Through this inclusion, physical channels are capable of providing consumers with more favorable experience and interaction, and brick-and-mortar store owners can obtain a more accurate understanding of consumer behaviors of store consumers. Information acquired through this system can be provided to store owners to serve as reference for merchandise placement, arrangement of display shelves, and consumer circulation path planning. This study used the technique of image processing to locate the Region of Interest and applied object tracking to get the consumer's trajectory which successfully implemented the consumer-tracking characteristics of online platforms in a physical channel while retaining the unique experience of the physical channel. This results in a win–win scenario for businesses and consumers. [ABSTRACT FROM AUTHOR]

In this paper, we propose an improved version of the neighbor embedding super-resolution (SR) algorithm proposed by Chang et al. [Super-resolution through neighbor embedding, in Proc. 2004 IEEE Computer Society Conf. Computer Vision and Pattern Recognition( CVPR), Vol. 1 (2004), pp. 275-282]. The neighbor embedding SR algorithm requires intensive computational time when finding the K nearest neighbors for the input patch in a huge set of training samples. We tackle this problem by clustering the training sample into a number of clusters, with which we first find for the input patch the nearest cluster center, and then find the K nearest neighbors in the corresponding cluster. In contrast to Chang's method, which uses Euclidean distance to find the K nearest neighbors of a low-resolution patch, we define a similarity function and use that to find the K most similar neighbors of a low-resolution patch. We then use local linear embedding (LLE) [S. T. Roweis and L. K. Saul, Nonlinear dimensionality reduction by locally linear embedding, Science 290(5500) (2000) 2323-2326] to find optimal coefficients, with which the linear combination of the K most similar neighbors best approaches the input patch. These coefficients are then used to form a linear combination of the K high-frequency patches corresponding to the K respective low-resolution patches (or the K most similar neighbors). The resulting high-frequency patch is then added to the enlarged (or up-sampled) version of the input patch. Experimental results show that the proposed clustering scheme efficiently reduces computational time without significantly affecting the performance. [ABSTRACT FROM AUTHOR]

This paper proposes an image effect creation system that can create various types of effects for images, to satisfy different needs of users for specific design goals. Some operations needed for creating image effects, such as flow-based bilateral filter, flow-based Gaussian filter, curve-shaped filters, line drawing, pencil texture generator, and modified shock filter, are first proposed, and various image effects that can be converted by the proposed system are illustrated. The experiments show very rich results in terms of image effects. [ABSTRACT FROM AUTHOR]