<title>Gray-scale-image-based character recognition algorithm for low-quality and low-resolution images</title>

Image enhancement and reconstruction is an important field of research in digital image analysis. To increase the quality of low-contrast images, a variety of image-enhancing technologies are available. Among these image enhancement techniques, singular value decomposition and discrete wavelet transform are the popular approaches for enhancing low-quality images. In this study, we have developed a modified singular value decomposition approach to enhance low-contrast and low-resolution close-circuit television images. Low-quality and low-resolution images with singular values near zero are used as input in the proposed approach. On the selected data, a threshold value was determined in the singular values of the diagonal matrix. Finally, the proposed modified singular value decomposition technique was applied to enhance the value of images with low quality and resolution. The datasets include 90 facial images of different individuals with low quality and resolution from a university database. The performance of the proposed approach is assessed using image entropy, peak signal to noise ratio, mean square error, contrast measurement and time computation. During the experiments performed on the obtained dataset, it was found that the proposed modified singular value decomposition approach outperformed the existing singular value decomposition, discrete wavelet transform-singular value decomposition and stationary wavelet transform methods.

Facial makeup style transfer is an extremely challenging sub-field of image-to-image-translation. Due to this difficulty, state-of-the-art results are mostly reliant on the Face Parsing Algorithm, which segments a face into parts in order to easily extract makeup features. However, we find that this algorithm can only work well on high-definition images where facial features can be accurately extracted. Faces in many real-world photos, such as those including a large background or multiple people, are typically of low-resolution, which considerably hinders state-of-the-art algorithms. In this paper, we propose an end-to-end holistic approach to effectively transfer makeup styles between two low-resolution images. The idea is built upon a novel weighted multi-scale spatial attention module, which identifies salient pixel regions on low-resolution images in multiple scales, and uses channel attention to determine the most effective attention map. This design provides two benefits: low-resolution images are usually blurry to different extents, so a multi-scale architecture can select the most effective convolution kernel size to implement spatial attention; makeup is applied on both a macro-level (foundation, fake tan) and a micro-level (eyeliner, lipstick) so different scales can excel in extracting different makeup features. We develop an Augmented CycleGAN network that embeds our attention modules at selected layers to most effectively transfer makeup. We test our system with the FBD data set, which consists of many low-resolution facial images, and demonstrates that it outperforms state-of-the-art methods, particularly in transferring makeup for blurry images and partially occluded images.

An automatic blood vessel segmentation is considered as important due to the presence of similar properties between blood vessels and the early stage pathologies (i.e. Microaneurysm and hemorrhages) in retinal diseases [1,2]. Blood vessel segmentation is considered as a challenging task, if a retinal images having inadequate contrast, lightning variations, and different pathologies like exudates. In literature, there is a number of blood vessel segmentation algorithms exist, but they are not compared on similar datasets and thus, it is difficult to choose which one is best. In this paper, we have implemented five different blood vessel segmentation techniques based on Gaussian matched filter [4], edge based techniques [5,8] (Sobel, Prewitt and Kirsch) and entropy based thresholding [6] technique and compare them on similar datasets with low quality and pathological retinal images. Secondly, in low quality images, most of the pixels have lost their color intensity information and considered as noise. In order to retrieve their color information of pixels, initial pre-processing is needed. Thus, a different pre-processing approach includes mask generation and contrast enhancement to retrieve the pixel information are proposed in retinal images and compared. The results obtained are verified broadly on two publicly available standard datasets DIARETDB0 and STARE dataset comprises of 211 retinal images.

Zoom based image super-resolution using DCT with LBP as characteristic model

To date, most optical character recognition (OCR) systems process binary document images, and the quality of the input image strongly affects their performance. Since a binarization process is inherently lossy, different algorithms typically produce different binary images from the same gray scale image. The objective of this research is to study effects of global binarization algorithms on the performance of OCR systems. Several binarization methods were examined: the best fixed threshold value for the data set, the ideal histogram method, and Otsu's algorithm. Four contemporary OCR systems and 50 hard copy pages containing 91,649 characters were used in the experiments. These pages were digitized at 300 dpi and 8 bits/pixel, and 36 different threshold values (ranging from 59 to 199 in increments of 4) were used. The resulting 1,800 binary images were processed by all four OCR systems. All systems made approximately 40% more errors from images generated by Otsu's method than those of the ideal histogram method. Two of the systems made approximately the same number of errors from images generated by the best fixed threshold value and Otsu's method.

The digitization of Tibetan ancient books is of great significance to the preservation of Tibetan culture. This problem involves two tasks: Tibetan text detection and Tibetan text recognition. The former is undoubtedly crucial to automatic Tibetan text recognition. However, there are few works on Tibetan text detection, and lack of training data has always been a problem, especially for deep learning methods which require massive training data. In this paper, we introduce the TxTAB dataset for evaluating text detection methods in Tibetan ancient books. The dataset is established based upon 202 treasured handwritten ancient Tibetan text images and is densely annotated with a multi-point annotation method without limiting the number of points. This is a challenging dataset with good diversity. It contains blurred images, gray and color images, the text of different sizes, the text of different handwriting styles, etc. An extensive experimental evaluation of 3 state-of-the-art text detection algorithms on TxTAB is presented with detailed analysis, and the results demonstrate that there is still a big room for improvements particularly for detecting Tibetan text in images of low quality.

The quality of character image plays an important role for the performance of character recognition system. However there is no good way to measure the recognition difficulty of a given character image. For the given character image with unknown quality, it is improper that apply the single character database to recognize it by the same feature and the same classifier. This paper proposed a novel approach for multi-resolution character recognition whose feature is extracted directly from gray-scale image and classification is adaptive classification which adaptively selects the appropriate character database and classifiers by evaluating the image quality of the input character. A resolution evaluation algorithm based on gray distribution feature was proposed to decide the adaptive classification weights for the classifiers, which make the classification have the higher probability of being the correct decision. Experiment results demonstrate the proposed approach highly improved the performance of character recognition system.

Eye localization is an important part in face recognition system, because its precision closely affects the performance of face recognition. Although various methods have already achieved high precision on the face images with high quality, their precision will drop on low quality images. In this paper, we propose a robust eye localization method for low quality face images to improve the eye detection rate and localization precision. First, we propose a probabilistic cascade (P-Cascade) framework, in which we reformulate the traditional cascade classifier in a probabilistic way. The P-Cascade can give chance to each image patch contributing to the final result, regardless the patch is accepted or rejected by the cascade. Second, we propose two extensions to further improve the robustness and precision in the P-Cascade framework. There are: (1) extending feature set, and (2) stacking two classifiers in multiple scales. Extensive experiments on JAFFE, BioID, LFW and a self-collected video surveillance database show that our method is comparable to state-of-the-art methods on high quality images and can work well on low quality images. This work supplies a solid base for face recognition applications under unconstrained or surveillance environments.

This paper introduces a spatio-temporal technique for selecting or filtering out lower quality digital image frames. The technique is demonstrated on Electro-Optical/Infrared image sequences which suggests it is a candidate for exploiting reconnaissance (recce) imagery or can be a part of a recce subsystem. For human vision exploitation, a few poor quality image frames out of hundreds in a digital image sequence may be only a minor irritation when the sequence runs at the typical 30 frames per se cond. Of course, if that human needs to examine each frame, a system that automatically removes or enhances lower quality image frames could be beneficial. For machine vision subsystems, a few poor quality image frames could cause lower probability of recognition.The filter technique introduced in this paper can improve input into machine vision algorithms. Another application for this technique is digital transmission to filter out unwanted images prior to transmission or to selectively enhance the poor quality frames. A major portion of current research into quality in digital image sequences focuses on transmission systems where an input high quality image sequence can be compared to the lower quality image sequence receivved at the output of the transmission system. However, this paper shows a technique for juding the quality of the input image frames prior to transmission, without a transmission system or without any knowledge of the higher quality image input. The impact of digital image artifacts on the spatio-temporal quality are shown. The quality variations in the individual frames of the input image sequence are charted to show which frames are of lower quality and thus need filtering.

Image Processing is any form of signal processing for which the image is an input such as a photograph or video frame. The output of image processing may be either an image or a set of characteristics or parameters related to the image. In many facial analysis systems like Face Recognition face is used as an important biometric. Facial analysis systems need High Resolution images for their processing. The video obtained from inexpensive surveillance cameras are of poor quality. Processing of poor quality images leads to unexpected results. To detect face images from a video captured by inexpensive surveillance cameras, we will use AdaBoost algorithm. If we feed those detected face images having low resolution and low quality to face recognition systems they will produce some unstable and erroneous results. Because these systems have problem working with low resolution images. Hence we need a method to bridge the gap between on one hand low- resolution and low-quality images and on the other hand facial analysis systems. Our approach is to use a Reconstruction Based Super Resolution method. In Reconstruction Based Super Resolution method we will generate a face-log containing images of similar frontal faces of the highest possible quality using head pose estimation technique. Then, we use a Learning Based Super-Resolution algorithm applied to the result of the reconstruction-based part to improve the quality by another factor of two. Hence the total system quality factor will be improved by four.

Character recognition and classification is considered one of the most important parts of current LPR systems. Because of low recognition quality and poor robustness of traditional character recognition techniques, those techniques were gradually replaced by powerful deep learning modules such as convolutional neural networks. Convolutional neural networks (CNNs) show satisfying ability in character recognition and outperform most of other available models. Since Chinese license plates contain Chinese characters in addition to ordinary alphanumeric characters, a robust, powerful, and efficient CNN is needed to accomplish character recognition task efficiently. In this paper, we have proposed an efficient CNN model based on Darknet architecture to perform character recognition. Through convolutional and max pooling layers, features of input character images will be extracted and then sent to softmax layer for classification. To avoid overfitting problem in the training process, the dropout regularization technique is adopted. We have used a dataset of 84,000 character images for training and testing our model. The experimental result shows satisfactory outputs and eventually achieves test accuracy of 99.69%.

Head pose estimation methods can be broadly classified into learning-based methods and model-based methods. The learning based methods use machine learning techniques to directly predict the pose from image appearance, while the model-based methods link the 2D observation (e.g. facial landmarks) and 3D model through the projection model for pose estimation. However, both methods may have difficulty on images with very low quality (e.g. low resolution, occlusion, and noisy images). For example, there would be limited appearance information to generate accurate landmark detection on lowquality images for reliable face pose estimation. To tackle pose estimation on low-quality images, we propose to combine the learning and model based methods. Specifically, we first build the relationship between facial landmark locations and image appearance using the Restricted Boltzmann Machine (RBM) model. Then, we link the landmark locations and 3D model analytically using the projection model. By combining the RBM model with the projection model, without explicit landmark detection, we predict the head pose with a KL-divergence based method and a gradient-based method. Experimental results demonstrate the effectiveness of the proposed method.

Secure transmission of images have become an important in case of its application and research. This study proposes a discrete wavelet-based resolution synthesis using interpolation in discrete wavelet transform (DWT) with a novel encryption method. Highly resoluted image has its importance in various applications like radar, electron microscopy, digital X-ray mammography, optical coherence tomography, astronomy, etc. High-resolution image is recovered from a low-resolution image where the image is transformed using Cohen-Daubechies-Feauveau 9/7 wavelet to obtain significant effectiveness in image resolution. By using the DWT, low-quality image is divided into various subbands of different frequency levels. These low-resoluted images are then interpolated using Cohen-Daubechies wavelet 9/7 coefficients and this interpolation provides nonlinearity over the image in order to preserve the visual smoothness edges. Preserved subband images are projected using fanbeam projection. These preserved various level images are used for reconstruction. High-resoluted image is then obtained using inverse discrete wavelet transform. These images are encrypted using the Pascal series. A significant improvement is obtained over other 2-D wavelet decomposition techniques such as bilinear, bicubic, nearest neighbor, and box wavelets. This study also implements this technique with the Digital Imaging and Communications in Medicine (DICOM) images for the betterment of performance metrics like peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). These results are produced using MATLAB simulation software.

Robust text detection via multi-degree of sharpening and blurring

<p>Text recognition task involves recognizing the text from the natural image; it possesses various application, which aids information extraction through data mining from street view like images. Scene text recognition involves two stages i.e., text detection and text recognition, in the past several mechanisms has been proposed for accurate identification, these mechanisms are either traditional approach or deep learning-based. All the existing deep-learning methodology fails as this comprises character data and image data, further this research develops an optimal architecture fusion neural network (FNN) for text identification and recognition. FNN comprises several layers of convolutional neural network as well as recurrent neural network. Within FNN architecture convolutional layer is utilized for the feature extraction and recurrent layer is utilized for attaining the feature classification prediction. Further, an optimal training architecture is established for the enhancement of classification accuracy. Here Devanagari MLT-19 dataset is utilized for the evaluation of FNN. Three different parameters are considered during evaluation i.e., script word identification, character recognition rate (CRR) and word recognition rate (WRR). Further comparison with existing models is performed to establish the proposed model efficiency and it shows FNN methodology observes 98.67% of script identification accuracy, 84.65% of WRR and 92.93% of CRR.</p>