Effective denoising is fundamental in image restoration, significantly impacting downstream computer vision tasks. Conventional CNN-based denoising models rely on paired training data of clean and noisy images, yet clean images are often unavailable in practical settings. Recent advancements in blind denoising, such as Noise2Void (N2V) and blind-spot networks, enable training without clean images; however, denoising quality remains an area for improvement. This paper introduces a novel Dense-in-Dense Network with Attention (DiDNA) designed specifically for blind denoising. By leveraging dense connections within a dense architecture and an attention module, DiDNA effectively captures complex noise patterns and enhances denoising capability. Experimental evaluations demonstrate that DiDNA not only surpasses existing blind denoising methods but also achieves competitive performance with traditional paired denoisers across CNN-based and non-CNN-based approaches.

In this paper, we propose an indoor action detection system which can automatically keep the log of users' activities of daily life since each activity generally consists of a number of actions. The hardware setting here adopts top-view depth cameras which makes our system less privacy sensitive and less annoying to the users, too. We regard the series of images of an action as a set of key-poses in images of the interested user which are arranged in a certain temporal order and use the latent SVM framework to jointly learn the appearance of the key-poses and the temporal locations of the key-poses. In this work, two kinds of features are proposed. The first is the histogram of depth difference value which can encode the shape of the human poses. The second is the location-signified feature which can capture the spatial relations among the person, floor, and other static objects. Moreover, we find that some incorrect detection results of certain type of action are usually associated with another certain type of action. Therefore, we design an algorithm that tries to automatically discover the action pairs which are the most difficult to be differentiable, and suppress the incorrect detection outcomes. To validate our system, experiments have been conducted, and the experimental results have shown effectiveness and robustness of our proposed method.

In this paper we propose a new data coding scheme called color intersection coding (CIC). The feature of the CIC is that people may not perceive particular color arrangement in encoded image at a distance, and he/she can shoot encoded image to decode hidden data at close distance. Color intersection is a key technique to generate four sub-pixels to represent a target pixel in original image, and the average color of those sub-pixels is equal to that of the target pixel. During decoding process, shot image is divided into several non-overlapping blocks, and then the average color of every block is computed. Color arrangement of four adjacent blocks is determined as well as hidden data is decoded. The experiment results demonstrate that the CIC scheme not only can control color contrast of encoded image, but also preserves content completeness of encoded image.

The high development of medicine causes the world's population aging quickly. To resolve the problem with limited medical resources, constant monitoring of elders' activity of daily living is important. We propose an activity recognition system for smart home, so elders can live alone and their children can monitor their parents' living activity to achieve the concept of "Aging in Place". The living activity monitoring model is powerful to recognize meaningful activities by using both ambient and wearable sensors. It's feasible to deploy in the real living environment be-cause it's a non-parametric learning model. Elders need less effort to label activity in training part, and the model may have chance to find some special activities that the elders did not consider in the past. We demonstrate the living activity monitoring model is feasible to be deployed in a living home with high accuracy performance of the activity recognition result.

With the popularity of Internet of Things, lots of resource constrained devices equipped with sensors and actuators are pervasively deployed to compose a smart environment, and Big Data are obtainable for a system to do further analytics thus to achieve human-centric purposes. One such human-centric system is a smart home which analyze Big Data to recognize contexts and their corresponding preferences for service configuration thus to provide context-aware services. However, since these Big Data are generated in real-time with huge amount, analytics based on conventional supervised way is not desirable due to the requirement of human efforts. In addition, there are usually multiple inhabitants with multiple combination of contexts in a home environment, and it is difficult to fully collect all these possible context combination as well as their corresponding preferences in advance. Therefore, this paper proposes an unsupervised nonparametric analytics method with a framework for human-centric smart homes to automatically discover contexts and their corresponding service configurations, and the models resulting from the proposed analytics can also be used to determine the preference for a context combination unseen before.

In this paper we propose a multi-scale image sharing scheme to hide multiple images to two meaningful sharing images. The proposed method combines conventional 2-out-of-2 visual cryptography with data hiding technique. The overall effort of the proposed scheme is the achievement of decrypting/extracting multiple secret images and reference images from sharing images at different scale-levels. Moreover, it is not only applied to monochromatic images but also color halftone images. The experiment results will demonstrate that the proposed scheme indeed increases more capacity of hidden images than the conventional 2-out-of-2 visual cryptography does.

The objective of this work is to propose a new template matching scheme which is able to deal with the recognition issue against rotation. The proposed scheme, rotation-invariant filter-driven template matching (RI-FTM), starts to transform a Cartesian-coordinate pattern to a polar-coordinate pattern. Subsequently, we put our emphasis on how to estimate an appropriate filter which is adopted to establish the connection between query pattern and reference pattern, and then the similarity between those two patterns is computed via sum of squared differences (SSD). In addition, the proposed method can shorten the width of filter to reduce the computing time. The experiment results will demonstrate our method is capable of recognitions of license plate characters and commercial logos.

The objective of this research is to design a new JPEG-based compression scheme which simultaneously considers the security issue. Our method starts from dividing image into non-overlapping blocks with size 8×8. Among these blocks, some are used as reference blocks and the rest are used as query blocks. A query block is the combination of the residual and the resultant of a filtered reference block. We put our emphasis on how to estimate an appropriate filter and then use it as part of a secret key. With both reference blocks and the residuals of query blocks, one is able to encode secured images using a correct secret key. The experiment results will demonstrate that how different secret keys can control the quality of restored image based on the priority of authority.

In this paper, we propose a new scheme to recognize old-film scene and modern scene. The proposed method automatically divides a video into several scenes. An adaptive histogram transform is performed on frames to yield flicker-free frames, and intensity flicker is measured and defined as the difference between original frame and flicker-free frame. A quantitative measurement of intensity flicker is computed as feature. Moreover, using color feature and the results of scene recognition, the proposed method identifies the characteristic of video and estimates the era of film in three periods, including, early period, middle period, and modern period. The experiment results will demonstrate that the proposed method not only recognizes old-film scene and modern scene, but it can handle video identification and era estimation of film and video.

This paper presents a novel people counting system for an environment in which a stationary camera can count the number of people watching a TV-wall advertisement or an electronic billboard without counting the repetitions in video streams in real time. The people actually watching an advertisement are identified via frontal face detection techniques. To count the number of people precisely, a complementary set of features is extracted from the torso of a human subject, as that part of the body contains relatively richer information than the face. In addition, for conducting robust people recognition, an online classifier trained by Fisher's Linear Discriminant (FLD) strategy is developed. Our experiment results demonstrate the efficacy of the proposed system for the people counting task.

This paper presents the adaptive histogram adjustment (AHA) to improve image contrast. The proposed method is achieved based on the concepts of weighted histogram separation and gray-level grouping. It not only improves the contrast of the local detail but also solves group density and blocky effect which occur at the under-quantization problem of weighted histogram separation. Moreover, the adaptive histogram adjustment can prevent the contrast over-enhancement of conventional adaptive-based approach. The experimental results show that AHA has good contrast sensitivity, and it is compared with the five existing approaches, such as, histogram equalization, adaptive histogram equalization, weighted histogram separation, adaptive weighted histogram separation and gray-level grouping.

We propose a new algorithm to improve the visual appearance of compressed videos, which are derived and reproduced from old films. The new technique is especially useful in handling the digitized old films of digital archives. Two main techniques, saturation adjustment and contrast enhancement, are implemented to improve the chrominance and luminance of video frames on HSL color space. For contrast enhancement, we use weighted histogram separation (WHS) to improve the luminance. Subsequently, the saturation ratio transfer function is proposed to adjust the saturation level of every frame. Furthermore, the new calculation method for the error of inter-frame is adopted to avoid the incorrectly enhanced result and the intensive computation consumed in motion estimation. The experimental results demonstrate that the proposed method outperforms the existing approaches in terms of video quality.

This paper proposes a novel automatic video diagnosing method. The purpose of this study is to detect the attack added on the authorized video and identify the attack category. The video is added with the crypto-watermarks using dual-domain quaternary watermarking algorithm in advance. The crypto-watermarks, which are generated using visual cryptography, have different capabilities against the various attacks. We extract the watermarks from the suspected video, measure the bit-error-rate between the extracted and specified crypto-watermarks, and then analyze bit-error-rates to determine what kind of attack added on the video. The experimental results demonstrate that the proposed method not only can identify the single attack, but it can identify the composite attack and detect the corrupted frames. Even if the video is not embedded with crypto-watermarks, we can differentiate it from the authorized videos.

This paper presents a method to localize and to recover the tampered areas in the watermarked image simultaneously. First, the proposed watermarking algorithm aims to embed a bi-watermark, which is the binary halftone of the downscaled host image, into the host image. While the watermarked image is tampered unauthorizedly, the extracted bi-watermark can be exploited to localize the tampered areas automatically without initial bi-watermark. Subsequently, Gaussian lowpass filter and quadratic programming are used to restore the bi-watermark to gray-scale image for inverse halftoning procedure. Eventually, the restored gray-scale image is used to recover the tampered areas.

A 2.5-domain tri-watermarking algorithm (2.5D-TW) is presented in this paper. The proposed algorithm is the integration of dual domain bi-watermarking algorithm and visual cryptography, and the embedded tri-watermark is composed of three binary watermarks. The tri-watermark is embedded into the encoded blocks in discrete cosine transform (DCT) domain, however, the tri-watermark can be extracted from the DCT blocks and the decoded frames in spatial domain. Moreover, two types of third binary watermark of the tri-watermark are developed namely the intra-watermark and inter-watermark, those binary watermarks reveal the different robustness against various attacks. The proposed method attempts to localize the tampered areas and to detect the temporal attack.

We propose a novel semi-fragile multiple-watermarking algorithm based on quantization index modulation. This algorithm utilizes two quantization steps to yield the non-uniform intervals in the real-number axis. Each interval corresponds to one binary symbol, includes stable-zero (S0), unstable-zero (U0), stable-one (S1), and unstable-one (U1). In addition, visual cryptography is integrated with the watermarking algorithm to increase the watermark capacity. Therefore, the host image is embedded the multiple watermarks, and then we extract the watermarks from the corrupted image. According to the extracted watermarks, the algorithm achieves the tamper proofing and attack identification. From the experimental result, it shows single and multiple tampered areas are detected and demonstrates that the amount of test images will not influence the accuracy of attack identification.

This paper presents a modified approach to the successive mean quantization transform, which is called as the weighted histogram separation (WHS) for enhancement of color images. Property of WHS situates between the successive mean quantization transform and the histogram equalization. In addition, this approach is further applied to the local enhancement, which is similar to the adaptive histogram equalization, and it is termed as the adaptive weighted histogram separation (AWHS). A comparison with successive mean quantization transform and histogram equalization has been performed in the experiments.

A dual-domain bi-watermarking algorithm is proposed in this paper. This algorithm embeds bi-watermark in DCT domain, but bi-watermark can be extracted in both spatial and DCT domains. Therefore, it can implement on the DCT-based compressed image/frame. The dual-domain bi-watermarking algorithm is the extension of the spatial-domain bi-watermarking algorithm, which implements quantization index modulation with two quantization step sizes. These quantization step sizes construct the non-uniform quantization intervals. Additionally, the luminance quantization table of JPEG compression is considered in the algorithm. In the experimental result, two extracted watermarks show the capability for various compression rates, and the watermarks also reveal the different robustness against the global and the regional attacks.

A novel image recovery algorithm for removing visible watermarks is presented. Independent component analysis (ICA) is utilized to separate source images from watermarked and reference images. Three independent component analysis approaches are examined in the proposed algorithm, which includes joint approximate diagonalization of eigenmatrices, second-order blind identification, and FastICA. Moreover, five different visible watermarking methods to embed uniform and linear-gradient watermarks are implemented. The experimental results show that visible watermarks are successfully removed, and that the proposed algorithm is independent of both the adopted ICA approach and the visible watermarking method. In the final experiment, several public domain images sourced from various websites are tested. The results of this study demonstrate that the proposed algorithm can blindly and successfully remove the visible watermarks without knowing the watermarking methods in advance

This work presents a novel algorithm using color contrast enhancement and lacuna texture synthesis is proposed for the virtual restoration of ancient Chinese paintings. Color contrast enhancement based on saturation and de-saturation is performed in the u'v'Y color space, to change the saturation value in the chromaticity diagram, and adaptive histogram equalization then is adopted to adjust the luminance component. Additionally, this work presents a new patching method using the Markov Random Field (MRF) model of texture synthesis. Eliminating undesirable aged painting patterns, such as stains, crevices, and artifacts, and then filling the lacuna regions with the appropriate textures is simple and efficient. The synthesization procedure integrates three key approaches, weighted mask, annular scan and auxiliary, with neighborhood searching. These approaches can maintain a complete shape and prevent edge disconnection in the final results. Moreover, the boundary between original and synthesized paintings is seamless, and unable to distinguish in which the undesirable pattern appears.