نشریه ماشین بینایی و پردازش تصویر
سال هشتم شماره 1 (بهار 1400)

Today, it is very important to authentication and identification in different organizations; therefore, the presentation of authentication systems with the ability to identify individuals and matching the input pattern has received much attention. The offline signature verification system is one of the biometric subsets of behavior used to verify the identity of the claimant. One of the major challenges in the signature feature is the reduction of intra-class diversity among the genuine and forged samples. For this purpose, in this paper, in order to increase the performance of the system, the combination of parametric features such as extraction of the feature in the radius of the intersecting points and the local binary pattern is discussed. In the proposed method, considering the spatial distribution of connected pixels, the neighbors of the candidate points are examined; therefore, by having local details about the candidate points, the strokes, arcs and angles of the signature pixels are extracted. The experiments used standard MCYT, GPDSsynthetic and CEDAR databases. The samples were separated using the KNN binder and based on the writer independent structure. According to the results, the average error rate in each database is 0.036, 0.033 and 0.12, respectively. In addition, the results of sensitivity and specificity criteria have improved compared to previous work.

Small size multi-spectral cameras are generating recently due to the fast development in UAVs technology and their application in agricultural field. These cameras are designed based on multi lenses structure where each lens captures images in special electromagnetic waves. There is a challenge in band to band co-registration due to this multi lens structure. On the other hand, Low altitude flight is necessary for some applications like plant disease detection where sub-pixel accuracy is needed to detect small features. Therefore, there is a higher miss-registration problem based on relief displacement in low flight altitude on top of trees after registration. The purpose of this article is to reduce the relief displacement error by introducing the most efficient image registration method. Therefore, three different datasets with different altitude are applied for this purpose. Results showed the proposed method was successful %83 to reduce miss-registration error.

Shape detection in digital images is one of the most effective subjects in image processing. This paper introduces a method for circle detection in digital images using genetic algorithm. Circle is expressed by a quadratic relation in coordinates screen. For circle detection, finding coefficients of this quadratic relation is the challenge. In the proposed method, three random points are selected on the edges of image. Because of each three points that do not locate on a straight line express a circle in coordinates screen, the quadratic relation coefficients of the circle are considered as the chromosome of genetic Algorithm. After finding the coefficients and drawing the circle, fitness function is calculated by computing amount of overlapping this circle with the edge of image. Then, the polynomial coefficients of new generations are generated by using crossover and mutation operators. Genetic algorithm continues until reaching the final conditions. Results of experiments on some of the images show that the proposed method can find circles on images. Increasing of the success rate in circle detection on image and exact detection of center and radius of circle are the contribution of this article in comparison with studied methods.

Images compression is an inevitable part of almost any images processing system, including face recognition systems. One of the main challenges in face recognition systems is reduction of recognition ratio due to the lossy compression of the images.In this paper, a new approach for face images compression improve is presented  by producing new quantization tables in JPEG method, using metaheuristic algorithms. The criterion for selecting the best quantization tables is the recognition rate of the compressed images. The new tables not only do not reduce the recognition rate, but also have the ability to increase the compression ratio at the same time. Experiments have been performed at different intervals of the compression ratio by adjusting the quality parameter on different sets of the FERET database. The results of the studies indicate that the recognition rate is maintained or in some cases is even increased, despite the increase in the compression rate.

By considering the importance of detecting fish and other aquatic objects, and also by considering the specifications of marine environment such as low quality and brightness of underwater images, light break, high level of noise, the overlapping of the fish object with the plants and the sands, nowadays researchers are looking for suitable methods for separating and segmenting fishes in this category of images. So far, various algorithms have been proposed to separate the fish objects from the underwater images, but these algorithms often faced various challenges such as strong noise in underwater images and superwised-based segmentation algorithms for underwater images which caused to use assumptions as identify the approximate foreground from background region. In this research, a new method has been introduced for isolating the fish object from underwater images. In this approach, a new color space was created for more accurate separation of pixels and detection of the fish object by combining the components of both color spaces RGB and CIE L×a×b. The new proposed method eliminates the limitation of the famous salient object detection algorithm which necessarily considers an approximate regions of the foreground and background. The performance of the proposed method significantly has been investigated by using the common evaluation criteria including accuracy, recall, precision and F_measure, which these indexes have been increased 5%, 3%, 4% and 1%, respectively, compared to the best rival method.

Detecting objects most focused on by human eyes when viewing a scene is of great interest to the computer vision community. Although a large amount of detection algorithms are available, due to variety and complexity of the image structures, the obtained saliency maps are still not satisfying enough. In this paper, we have proposed an efficient, supervised algorithm for saliency map detection which uses a conditional random field. Integrating different salient cues with matrix decomposition methods through CRF is one of the innovations of this paper. Another achievement of this paper is considering potential weights, obtained from CRF training process, as a ranking tool to select the best saliency cues. Since CRF is a supervised method, some papers select, for training step, a number of images which are most similar to the input image. The present paper offers, as our third contribution, a comprehensive assessment of the methods which select such similar images. Evaluating the proposed method on the ECSSD and MSRA-10k datasets with respect to the evaluation criteria has indicated its excellent performance.