MACHINE VISION AND IMAGE PROCESSING
Year:2019 | Volume:6 | Issue:1|Papers Count:6

Today, with the development of technologies to capture and share images, the number of digital images has increased significantly. The management of this volume of images requires an efficient system to review, classify, search and retrieve the images. New generations of image retrieval systems usually take one or a few keywords from the user to retrieve images with visual content related to that keywords. A mechanism that can automatically describe the content of an image (like a human) can increase the efficiency of these systems. Automatic Image Annotation or AIA is a professional method to express the content of images by keywords or tags. AIA systems, investigate the relationship between the meaning of a text and low-level image features by using machine learning techniques. They automatically assign some tags to images to facilitate fast search based on image contents. In this paper, we explain the different steps to implement an AIA system, review the related works and express the problems and challenges in designing such systems. We also introduce several datasets suitable for AIA systems.

Principal component analysis (PCA) is one of the proposed methods to reduce the size of the data set that can be used for both one and two-dimensional data. Regarding the lack of sparsity property in the base vectors, sparse PCA has been proposed, which maintains the properties of standard PCA and simultaneously forces some of the elements of the base vectors to zero. In this paper, due to the sparsity in base vectors that cause some dataset values to be ineffective in moving to new space, two algorithms are presented in one-dimensional and two-dimensional mode to remove redundancy from raw data. In the one-dimensional algorithm, redundancy is detected between signal layers and then removed from all set observations. In a two-dimensional algorithm, the significance of the row and the column of the dataset images are detected and the less important ones are eliminated directly from raw data. One of the most important advantages of proposed algorithms, which can be read as non-uniform sampling methods, is to preserve the appearance of signals. After removing the raw data redundancy by the two algorithms presented, new data with fewer dimensions can be used in other applications such as dataset recognition, compression, and so on.

In this paper, a method based on deep learning is presented to highlight and recognize the Iranian license plate numbers. The current research uses the convolutional neural network with the encoder-decoder structure to enhance the image and highlight the plate image numbers instead of using traditional image enhancement techniques. The proposed network can highlight vehicle license plate numbers by learning the plate images in various conditions. After that, the plate numbers are recognized from the reproduced image using a recurrent neural network without the need to plate image segmentation. This method can reduce the error caused by the license plate number segmentation. The proposed method reached the final recognition rate up to 94٫ 19 percent on a database with 4000 test images for recognizing the license plates which is acceptable in comparison to three recent methods.

Most intelligent underwater vehicles and remote control marine vehicles are usually equipped with optical cameras for underwater photography. However, due to water properties and its impurity, the quality of the images taken by these imaging equipment is not desirable, enough. Because, water reduces the light and by increasing the deep of water, the further light will diminish, which will result in the absorption of colors by water. Hence, image processing operations are crucial for underwater images. In this paper, a new method has been proposed to improve the contrast and quality of underwater images. In the proposed method, the two methods of contrast stretching and histogram equalization have been employed. In the histogram equalization phase, new methods have been introduced for thresholding and histogram clipping. The proposed method has been applied on a benchmark images dataset and the results have been compared with the results of common methods. The analysis of the proposed method results, compared to the Sathya et al., which is the best method compares to other rival methods, shows 40. 16% increase in the contrast of the images, in the best case.

Unusual behavior detection is critically important for visual surveillance. It is also a challenging research topic in computer vision. Although much effort has been devoted to tackle this problem, such detection task in a realistic and uncontrolled environment is still far from mature. The major difficulty lies in the ambiguous characteristic in differentiating normal and abnormal behaviors, whose definitions often vary according to the context of video's history. In this paper we propose a framework for detecting and locating abnormal activities in video sequences. The key aspect of our method is the pairing of the 2D and 3D spatial-temporal Convolutional Neural Networks (CNN) for anomaly detection in contiguous video frames. The Features from Accelerated Segment Test (FAST) detector has been used In order to increase the reliability in identifying the interest locations in entry frames of convolutional neural network model. These feature extracted only from volumes of moving pixels that reduce the computational costs. The architecture of CNN model allows us to extract spatial-temporal features that contain complicated motion features. We test our framework on popular benchmark dataset containing various human abnormal activities and situations. Evaluation results show that our method outperforms most of other methods and achieves a very competitive detection performance compared to state-of-the-art methods.

Periodic noise damages the visual quality of images by imposingrepetitive patterns to them. In this research work, we introduce anew method which is based on fuzzy systems for de-noising periodic noise. The position of a frequency coefficient in the origin shifted Fouriertransformed image and the current coefficient's amplitude to a localmedian ratio are used as inputs of the fuzzy system. The output of the fuzzy system will be a restorationmask. Weimplemented the proposed method and evaluated its performance againstsome images corrupted with periodic noise. The experimental results showan acceptable level of performance. Overall, this research implies thatthe procedure conducted by experts in the notch filter can be automatedby using a fuzzy system.