Search

Existing learning-based atmospheric particle-removal approaches such as those used for rainy and hazy images are designed with strong assumptions regarding spatial frequency, trajectory, and translucency. However, the removal of snow particles is more complicated because it possess the additional attributes of particle size and shape, and these attributes may vary within a single image. Currently, hand-crafted features are still the mainstream for snow removal, making significant generalization difficult to achieve. In response, we have designed a multistage network codenamed DesnowNet to in turn deal with the removal of translucent and opaque snow particles. We also differentiate snow into attributes of translucency and chromatic aberration for accurate estimation. Moreover, our approach individually estimates residual complements of the snow-free images to recover details obscured by opaque snow. Additionally, a multi-scale design is utilized throughout the entire network to model the diversity of snow. As demonstrated in experimental results, our approach outperforms state-of-the-art learning-based atmospheric phenomena removal methods and one semantic segmentation baseline on the proposed Snow100K dataset in both qualitative and quantitative comparisons. The results indicate our network would benefit applications involving computer vision and graphics.

Local image contrast enhancement methods extract more informative details comparing with the global methods. Yet, the traditional schemes require to access all of the pixels in a local region to obtain the cumulative distribution function (cdf) for further enhancement, i.e., adaptive histogram equalization (AHE), and thus the time complexity drastically increases accordingly when the block size increases. Consequently, some speed-up ingenuities such as block-wise strategy and simple proportionally averaging manner are employed to cope with this issue, yet serious blocking artifact or low contrast issues are accompanied. In this study, a local enhancement method, namely corrected parametric-oriented histogram equalization (CPOHE), is proposed to effectively yield enhanced results with high contrast and artifact-free texture using the concept of integral image, and further correct the accompanied distortion. As documented in the experimental results, the proposed method provides a high practical value for applied in visual perception, biometric, and tracking/detection applications.

The conventional dual-mode halftoning methods achieve high quality halftone patterns by easing smooth artifact and preserving image details. However, the boundaries between low- and high-frequency image regions still present undesired texture in some particular cases, which significantly degrades the visual quality. In this work, a multi-mode halftoning method is proposed to deal with the object map artifact and boundary artifact simultaneously. In addition, the new absorptance-frequency stacking constraint is also employed to solve the noisy textures of the halftone outputs. As documented in the experimental results, high quality halftone outputs can be obtained, proving that the proposed method can be a competitive candidate for electrophotographic printers.

Formerly, dimensionality reduction techniques are effective ways for extracting statistical significance of features from their original dimensions. However, the dimensionality reduction also induces an additional complexity burden which may encumber the real efficiency. In this paper, a technique is proposed for the reduction of the dimension of samples rather than the features in the former schemes, and it is able to additionally reduce the computational complexity of the applied systems during the reduction process. This method effectively reduces the redundancies of a sample, in particular for those objects which possess partially symmetric property, such as human face, pedestrian, and license plate. As demonstrated in the experiments, based upon the premises of faster speeds in training and detection by a factor of 4.06 and 1.24, respectively, similar accuracies to the ones without considering the proposed method are achieved. The performance verifies that the proposed technique can offer competitive practical values in pattern recognition related fields.

In this study, a high accuracy fingerprint classification method is proposed to enhance the performance in terms of efficiency for fingerprint recognition system. The recognition system has been considered as a reliable mechanism for criminal identification and forensic for its invariance property, yet the huge database is the key issue to make the system obtuse. In former works, the pre-classifying manner is an effective way to speed up the process, yet the accuracy of the classification dominates the further recognition rate and processing speed. In this paper, a rule-based fingerprint classification method is proposed, wherein the two features, including the types of singular points and the number of each type of point are adopted to distinguish different fingerprints. Moreover, when fingerprints are indistinguishable, the proposed Center-to-Delta Flow (CDF) and Balance Arm Flow (BAF) are catered for further classification. As documented in the experimental results, a good accuracy rate can be achieved, which endorses the effectiveness of the fingerprint classification scheme for the further fingerprint recognition system.

Moving object detection is an important and fundamental step for intelligent video surveillance systems because it provides a focus of attention for post-processing. A multilayer codebook-based background subtraction (MCBS) model is proposed for video sequences to detect moving objects. Combining the multilayer block-based strategy and the adaptive feature extraction from blocks of various sizes, the proposed method can remove most of the nonstationary (dynamic) background and significantly increase the processing efficiency. Moreover, the pixel-based classification is adopted for refining the results from the block-based background subtraction, which can further classify pixels as foreground, shadows, and highlights. As a result, the proposed scheme can provide a high precision and efficient processing speed to meet the requirements of real-time moving object detection.

This paper presents an improved hand tracking system using pixel-based hierarchical-feature AdaBoosting (PBHFA), skin color segmentation, and codebook (CB) background cancellation. The proposed PBH feature significantly reduces the training time by a factor of at least 1440 compared to the traditional Haar-like feature. Moreover, lower computation and high tracking accuracy are also provided simultaneously. Yet, one of the disadvantages of the PBHFA is the false positive which is the consequence of the appearance of complex background in positive samples. To effectively reduce the false positive rate, the skin color segmentation and the foreground detection by applying the CB model are catered for rejecting all of the candidates which are not hand targets. As documented in the experimental results, the proposed system can achieve promising results, and thus it can be considered as an effective candidate in handling practical applications which require hand postures.

This paper presents a hierarchical scheme with block-based and pixel-based codebooks for foreground detection. The codebook is mainly used to compress information to achieve a high efficient processing speed. In the block-based stage, 12 intensity values are employed to represent a block. The algorithm extends the concept of the block truncation coding, and thus it can further improve the processing efficiency by enjoying its low complexity advantage. In detail, the block-based stage can remove most of the backgrounds without reducing the true positive rate, yet it has low precision. To overcome this problem, the pixel-based stage is adopted to enhance the precision, which also can reduce the false positive rate. Moreover, the short-term information is employed to improve background updating for adaptive environments. As documented in the experimental results, the proposed algorithm can provide superior performance to that of the former related approaches.

License plate localization (LPL) and character segmentation (CS) play key roles in the license plate (LP) recognition system. In this paper, we dedicate ourselves to these two issues. In LPL, histogram equalization is employed to solve the low-contrast and dynamic-range problems; the texture properties, e.g., aspect ratio, and color similarity are used to locate the LP; and the Hough transform is adopted to correct the rotation problem. In CS, the hybrid binarization technique is proposed to effectively segment the characters in the dirt LP. The feedback self-learning procedure is also employed to adjust the parameters in the system. As documented in the experiments, good localization and segmentation results are achieved with the proposed algorithms.

Processing efficiency can be a key factor which dominates the value of a commercial product and its value-added application such as security. In this study, a computationally reduced halftoning-based watermarking is proposed. In encoder, the Efficient Direct Binary Search (EDBS) is employed to generate reference table to ensure the output is in halftone format. Subsequently, a number of optimized compressive tables with various texture angles are established for subsequent table lookup. In decoder, the Least-Mean-Square (LMS) enlarges the differences among those phenotypes of the embedded angles and the required number of dimensions for each angle. Finally, the naive Bayes classifier is employed to collect the possibility information for classifying various angles. As documented in the experimental results, good image quality and correct detect rate can be obtained simultaneously. Moreover, a high processing efficiency of 0.6 milliseconds for an image of size 512×512 can also be achieved, which can substantially increase the commercial competitive strength in printing market, in particular the security issue is well addressed.

This paper presents a new face gender recognition scheme by enjoying the benefit from the dot diffusion among weak classifiers in recognition phase for a low resolution and non-aligned thumbnail image. The main problem of the former Adaboost approaches is that each weak classifier simply offers a binary decision, which fails to compensate the decision error by diffusing it to the rest weak classifiers. To cope with this, this work exploits the dot-diffused-based Adaboost to solve this problem. As documented in the experimental results, with the examination of Feret and CMU databases, this paper has shown that the proposed scheme is an effective candidate in improving the recognition accuracy rate and the efficiency of the overall system process for face gender recognition.

In this paper, an improved system which can detect multiple license plates in high resolution applications is proposed. First, the Symmetric Mask-based Discrete Wavelet Transform (SMDWT) is utilized to speed up the license plates detection. Later on, the proposed Hierarchical AdaBoost (HA) realizes the detection for multiple targets, in which the classifiers simply require 10 minutes off-line training. In the character segmentation phrase, a projection-based approach considering the relationship among characters is also established. Finally, 13 structural/directional features with the Naive Bayes are applied for character recognition. Experimental results conclusively ensure that the proposed technique is effective for freeway's multiple toll station and transportation management applications contributed from excellent reliability and processing efficiency in big frame size.

In this paper, a halftoning-based watermarking method is presented. This method enables high pixel-depth watermark embedding, while maintaining high image quality. This technique is capable of embedding watermarks with pixel depths up to 3 bits without causing prominent degradation to the image quality. To achieve high image quality, the parallel oriented high-efficient direct binary search (DBS) halftoning is selected to be integrated with the proposed orientation modulation (OM) method. The OM method utilizes different halftone texture orientations to carry different watermark data. In the decoder, the least-mean-square-trained filters are applied for feature extraction from watermarked images in the frequency domain, and the naive Bayes classifier is used to analyze the extracted features and ultimately to decode the watermark data. Experimental results show that the DBS-based OM encoding method maintains a high degree of image quality and realizes the processing efficiency and robustness to be adapted in printing applications.

In this paper, a halftoning-based watermarking scheme with high data capacity and image quality is presented. Three types of watermarks of various pixel-depths, including 1-bit, 2-bit, and 3-bit, are able to be embedded without prominently damaging the image quality. To achieve high marked image quality, the parallel-oriented high efficient Direct Binary Search (DBS) halftoning is adopted to cooperate with the proposed Orientation Modulation (OM) method. In the decoder, the Least-Mean-Square-trained (LMS-trained) filters are adopted to extract the features of marked images in the frequency domain, and the naive Bayes classifier is employed to analyze the extracted features and further decode the watermark information. Experimental results demonstrate that the proposed DBS-based OM encoding scheme provides excellent image quality, high processing efficiency, and high robustness to adapt to practical printing application.

In this paper, a classified-based post-compensation algorithm for Color Filter Array (CFA) demosaicing is proposed. This technique can be used for improving the image quality of the interpolated results obtained by other CFA images. First, each pixel is classified according to its neighborhood texture variance and angle. Then, different Least-Mean-Square (LMS) filters are trained to adopt for dealing pixels of various characteristics. As documented in the experimental results, the proposed scheme can substantially boost the image quality; in addition, a better visual perceptual can be obtained. Notably, the proposed method can be considered as effective post-compensation by applying for any former schemes to yield an even better image quality.

This study proposes a method which can generate high quality inverse halftone images from halftone images. This method can be employed prior to any signal processing over a halftone image or the inverse halftoning used in JBIG2. The proposed method utilizes the least-mean-square (LMS) algorithm to establish a relationship between the current processing position and its corresponding neighboring positions in each type of halftone image, including direct binary search, error diffusion, dot diffusion, and ordered dithering. After which, a referenced region called a support region (SR) is used to extract features. The SR can be obtained by relabeling the LMS-trained filters with the order of importance. Moreover, the probability of black pixel occurrence is considered as a feature in this work. According to this feature, the probabilities of all possible grayscale values at the current processing position can be obtained by the Bayesian theorem. Consequently, the final output at this position is the grayscale value with the highest probability. Experimental results show that the proposed method offers better visual quality than that of Mese-Vaidyanathan's and Chang et al's methods in terms of human-visual peak signal-to-noise ratio (HPSNR). In addition, the memory consumption is also superior to Mese-Vaidyanathan's method.

In this work, a method which can generate high quality inverse halftone images from halftone images is proposed. This method uses least-mean-square (LMS) trained filters to establish the relationship between the current processing position and its corresponding neighbor positions in each kind of halftone image. This includes direction binary search (DBS), error diffusion, dot diffusion, and ordered dithering. After which, the support region which is used for features extracting can be obtained by relabeling the LMS-trained filters by order of importance. Two features are used in this work: 1) the probability of black pixel occurrence at each position in the support region, and 2) the probability of mean occurrence which is obtained from all pixels in the support region. According to these data, the probabilities of all possible grayscale values appearance at current processing position can be obtained by Bayesian theorem. Consequently, the final output at this position is the grayscale value with highest probability. Experimental results show that the image quality and memory consumption of the proposed method are superior to Mese-Vaidyanathan's method.

License Plate Localization (LPL) and Character Segmentation (CS) play key roles in License Plate Recognition System (LPRS). In this study, we dedicate ourselves in these two issues. In LPL, the histogram equalization is employed to solve the low contrast and dynamic range problem; the texture properties, e.g., aspect ratio, and color similarity are used to locate the License Plate (LP). In CS, the hybrid- binarization technique is proposed to effectively segment the characters in the dirt LP. The feedback self-learning procedure is also employed to adjust the parameters in the system. As documented in the experiments, good localization and segmentation results are achieved with the proposed algorithms.