MSER을 이용한 문서 이미지 이진화 기법

Extremal Regions of Extremum Levels (EREL) are regions detected from a set of all extremal regions of an image. Maximally Stable Extremal Regions (MSER) which is a novel affine covariant region detector, detects regions from a same set of extremal regions as well. Although MSER results in regions with almost high repeatability, it is heavily dependent on the union-find approach which is a fairly complicated algorithm, and should be completed sequentially. Furthermore, it detects regions with low repeatability under the blur transformations. The reason for the latter shortcoming is the absence of boundaries information in stability criterion. To tackle these problems we propose to employ prior information about boundaries of regions, which results in a novel region detector algorithm that not only outperforms MSER, but avoids the MSER's rather complicated steps of union-finding. To achieve that, we introduce Maxima of Gradient Magnitudes (MGMs) and use them to find handful of Extremum Levels (ELs). The chosen ELs are then scanned to detect their Extremal Regions (ER). The proposed algorithm which is called Extremal Regions of Extremum Levels (EREL) has been tested on the public benchmark dataset of Mikolajczyk [1]. Our experimental evaluations illustrate that, in many cases EREL achieves higher repeatability scores than MSER even for very low overlap errors.

Document image binarization has been a long standing problem for unconstrained document images. Although various thresholding algorithms have been developed over the years, problems associated with strong noise, complex patterns, poor contrast, and variable modalities in gray-scale histograms still limit the performance of document image analysis systems. Given the unpredictable nature of these image attributes, few thresholding algorithms work consistently well for document image binarization. This dissertation presents texture feature based thresholding algorithms to address these difficulties. The philosophy of our thresholding approach is that the texture domain knowledge of document images is important to judge the binarization quality and thus guide the binarization process; that is, suitably defined texture features of document images can be used to assist the optimal threshold selection. Our thresholding scheme consists of three steps. First, candidate thresholds are produced through the iterative use of Otsu's algorithm. Second, texture features associated with each candidate threshold are extracted from the run-length histogram of the accordingly binarized image. Third, the optimal threshold is selected so that the most desirable document texture features are preserved. This thresholding scheme was implemented in both global and adaptive modes. With our program design the algorithms require only one image scan pass, facilitating their hardware implementation for a commercial system. Experimental results with 9000 machine printed address blocks from an unconstrained US mail stream demonstrated that over 99.6% of the images were well binarized by our thresholding method, which are appreciably better than those obtained by existing thresholding techniques. Also a system run with 500 difficult mail address blocks showed that an 8.1% higher character recognition rate was achieved by our algorithm in comparison to that by Otsu's algorithm.

This paper presents the method to classify the speed limit traffic sign using multiple features, namely histogram of oriented gradient (HOG) and maximally stable extremal regions (MSER) features. The classification process is divided into the outer circular ring matching and the inner part matching. The HOG feature is employed to match the outer circular ring of the sign, while MSER feature is employed to extract the digit number in the inner part of the sign. Both features are extracted from the grayscale image. The algorithm detects the rotation angle of the sign by analyzing the blobs which is extracted using MSER. In the matching process, tested images are matched with the standard reference images by calculating the Euclidean distance. The experimental results show that the proposed method for matching the outer circular ring works properly to recognize the circular sign. Further, the digit number matching achieves the high classification rate of 93.67% for classifying the normal and rotated speed limit signs. The total execution time for classifying six types of speed limit sign is 10.75 ms.

In order to reduce the impact of shooting angle, distance and motion blur on vehicle number localization, and improve the accuracy of the localization in complex environments, a new vehicle number localization method based on edge-enhanced maximally stable extremal region (MSER) is proposed. This method first performs edge enhancement on the vehicle number image and extracts the character candidate region in the edge-enhanced image by using MSER. Then, the candidate region is filtered according to the vehicle number character features using the heuristic rule and the stroke width rule, and the remaining candidate regions are merged. The final step is completed by combining the geometric features of the vehicle number. The experimental results show that the comprehensive localization accuracy of this method is as high as 98.50%. High accuracy and robustness can be achieved by using this method in vehicle number localization with images acquired in complicated scenes with various shooting distances and angles.

In order to solve the problem of automation of cable number identification in the process of cable assembly of large equipment, we proposed a cable number detection and recognition method based on image Maximally Stable Extremal Regions (MSER) and convolutional neural network (CNN). We used image processing algorithms combined with deep learning to achieve end-to-end identification of cable number images. In view of the uncertain arrangement order and scattered distribution of numbers on the cable, the MSER algorithm was improved by image processing in order to extract the area containing numbers more accurately. The number identification problem in the case of lack of sample database is solved by the migration-trained CNN network. In addition, the recognition effect of the method under different environmental conditions has been studied. Through experimental research, the optimized MSER and CNN algorithms can better complete the cable number detection and identification tasks.

This paper introduces a tracking method for the well known local MSER (Maximally Stable Extremal Region) detector. The component tree is used as an efficient data structure, which allows the calculation of MSERs in quasi-linear time. It is demonstrated that the tree is able to manage the required data for tracking. We show that by means of MSER tracking the computational time for the detection of single MSERs can be improved by a factor of 4 to 10. Using a weighted feature vector for data association improves the tracking stability. Furthermore, the component tree enables backward tracking which further improves the robustness. The novel MSER tracking algorithm is evaluated on a variety of scenes. In addition, we demonstrate three different applications, tracking of license plates, faces and fibers in paper, showing in all three scenarios improved speed and stability.

In this paper, an affine invariance feature detection method based on Scale Invariant Feature Transform (SIFT) and Maximally Stable Extremal Regions (MSER) is proposed. Classical SIFT algorithm is not robust to affine deformations, because it is based on DOG detector which extracts circle regions for keypoint location. In order to overcome this disadvantage, DOG detector in conventional SIFT algorithm is replaced by MSER detector which is robust to affine deformation. Then these regions are normalized and extracted using SIFT. Simulation studies are carried out to show the effectiveness of the proposed method to affine transform in comparison to traditional SIFT algorithm.

In this paper, we introduce WignerMSER, a robust local feature detector for thermal infrared images, a new variant of renowned maximally stable extremal regions (MSER) feature detector. Thermal infrared images pose challenges of low contrast and lack of sharp edges or boundaries typically required for identifying keypoints using traditional feature detectors. This paper addresses this challenge by designing a local feature detector suitable for infrared images. The proposed WignerMSER features are obtained by transforming the image from the spatial domain to joint space-spatial-frequency domain using pseudo Wigner-Ville Distribution, and thereafter detecting MSERs in the Wigner transformed space. Different variants of WignerMSER are proposed and their performance is compared over traditional MSER feature detector in a bag of words framework for the purpose of object recognition in thermal infrared images. The performance of WignerMSER feature detector is evaluated on two infrared datasets. The first dataset is CSIR-CSIO thermal infrared (TIR) vehicle dataset comprising of four vehicle categories commonly seen on urban roads in India and the second dataset is a subset of FLIR Thermal dataset comprising of two object categories, vehicles, and pedestrians. All the variants of the proposed WignerMSER feature detector have demonstrated superior classification performance against traditional MSER features for object categorization on both the TIR datasets. Fused WignerMSER feature detector has shown the best performance with more than 5% increase in overall classification accuracy over traditional MSER for both the datasets.

Document image binarization plays an important role in image segmentation and its effect directly impacts on the quality of the OCR recognition system. However, binarization is difficult for camera-based document images with poor contrast or illumination. In this paper, we propose a binarization algorithm, called NFCM, for camera-based document image. NFCM, a local threshold method, is a combination of Niblack algorithm and FCM (Fuzzy C-Means) algorithm. It is good at not only preserving the character stokes, but also alleviating the ghost artifacts. Comparative experiments show that NFCM can obtain favorable results with respect to the OCR performance.

The visual analysis of foods on social media by using food recognition algorithm provides valuable insight from the health, cultural and marketing. Food recognition offers a means to automatically recognise foods as well the useful information such as calories and nutritional estimation by using image processing and machine learning technique. The interest points in food image can be detected effectively by using Maximally Stable Extremal Region (MSER). As MSER used global segmentation and many food images have a complex background, there are numerous irrelevant interest points are detected. These interest points are considered as noises that lead to computation burden in the overall recognition process. Therefore, this research proposes an Extremal Region Selection (ERS) algorithm to improve MSER detection by reducing the number of irrelevant extremal regions by using unsupervised learning based on the k-means algorithm. The performance of ERS algorithm is evaluated based on the classification performance metrics by using classification rate (CR), error rate (ERT), precision (Prec.) and recall (rec.) as well as the number of extremal regions produced by ERS. UECFOOD-100 and UNICT-FD1200 are the two food datasets used to benchmark the proposed algorithm. The results of this research have found that the ERS algorithm by using optimum parameters and thresholds, be able to reduce the number of extremal regions with sustained classification performance.

The natural scene images contain valuable information about themselves in the form of textual matter present in them. The process of extracting text regions can be used to understand the context of the image, that can be of great help in many applications helpful for humanity. The extraction of text regions from natural scene images, which is a daunting task due to variation in text elements in the form of size, orientation, colors, low contrast images and complicated background. In this paper, we propose a method based on Maximal Stable Extremal Region (MSER) and weighted median filter along with three text specific traits to identify and extract text regions by creating bounding box around them in natural scene images. The image is passed through a weighted median filter to preserve and smoothen the edges followed by candidate region extraction by MSER. Heuristics rules filter the non-text components. Finally, the classification process is carried out with the help of classifiers (using adaboost.m1 and k-nn) to classify candidate text regions and non-text regions based on three text specific traits, followed by the grouping of text components in text line using clustering. The method aims to extract text regions robustly from low contrast images. The performance of the method is checked on ICDAR 2011 testing dataset to prove its efficiency concerning precision, recall, and f-measure.

In this paper a novel robust algorithm that extracts text from signage images is proposed. Existing methods are constructed to work on certain limitations, as detecting text under perfect lightning conditions. Maximally Stable Extremal Regions (MSER) detector is used to extract text areas because it shows better performance to lightning conditions. To enhance the MSER regions, novel enhancement technique is used by combining edge detection algorithm. Enhance edge image is then put to the stroke width detection algorithm. Based on stroke width information, connected components are obtained. Afterwards connected components are filtered to remove non text pixels. Text recognition is not touched by this algorithm and resulted binary image can be used on any text recognition algorithm. This algorithm shows best performance on light effected text images and noisy images.

Emotion recognition has been one of the stimulating issues over the years due to the irregularities in the complexity of models and unpredictability between expression categories. So many Emotion detection algorithms have developed in the last two decades and still facing problems in accuracy, complexity and real-world implementation. In this paper, we propose two feature extraction techniques: Mouth region-based feature extraction and Maximally Stable Extremal Regions (MSER) method. In Mouth based feature extraction method mouth area is calculated and based on that value the emotions are classified. In the MSER method, the features are extracted by using connecting components and then the extracted features are given to a simple ANN for classification. Experimental results shows that the Mouth area based feature extraction method gives 86% accuracy and MSER based feature extraction method outperforms it by achieving 89% accuracy on DEAP. Thus, it can be concluded that the proposed methods can be effectively used for emotion detection.

Document image binarization is a central problem in many document analysis systems. Indeed, it represents one of the basic challenges, especially in case of historical documents analysis. In this paper, we propose a novel robust multi stage framework that combines different existing document image thresholding methods for the purpose of getting a better binarization result. CLAHE technique is introduced to significantly enhance contrast in some poor images. The proposed method then uses a hybrid algorithm to partition image into foreground and background. A special procedure is finally applied in order to remove small noise and correct characters morphology. Experimental results prove the accuracy and the efficiency of our approach on document images binarization over three popular datasets compared to some well-known methods in literature.

Food object recognition has gained popularity in recent years. This can perhaps be attributed to its potential applications in fields such as nutrition and fitness. Recognizing food images however is a challenging task since various foods come in many shapes and sizes. Besides having unexpected deformities and texture, food images are also captured in differing lighting conditions and camera viewpoints. From a computer vision perspective, using global image features to train a supervised classifier might be unsuitable due to the complex nature of the food images. Local features on the other hand seem the better alternative since they are able to capture minute intricacies such as interest points and other intricate information. In this paper, two local features namely SURF (Speeded- Up Robust Feature) and MSER (Maximally Stable Extremal Regions) are investigated for food object recognition. Both features are computationally inexpensive and have shown to be effective local descriptors for complex images. Specifically, each feature is firstly evaluated separately. This is followed by feature fusion to observe whether a combined representation could better represent food images. Experimental evaluations using a Support Vector Machine classifier shows that feature fusion generates better recognition accuracy at 86.6%.