نشریه ماشین بینایی و پردازش تصویر
سال هشتم شماره 3 (پاییز 1400)

Local Binary Pattern (LBP) is one of popular texture images descriptors. It is used to extraction features of texture images for classification.  However, in this paper a new version of LBP is proposed that is not used for this purpose.  The proposed descriptor is used for image retrieval. In other words it can be compare to Scale Invariant Features Transform (SIFT). In term of speed, the proposed method outperforms the SIFT. In addition, in this paper it is used for Infrared images retrieval. These images are low quality and low contrast and if they are used without any pre processing and enhancement, SIFT descriptor versions cannot extract good features of them. The proposed method can provide better result with higher speed. The proposed method have been compared to center symmetric LBP (CS-LBP) and SIFT by using both infrared images and standard image data-sets.

In this paper a new texture descriptor is presented for biometric authentication systems based on hand dorsal texture characteristics. First, different hand images are collected using an acquisition device, then the images are mapped from RGB color space to YUV and the five fingers are isolated from images with thresholding of the U component for skin detection. Utilizing the center of gravity, length and direction of the fingers, the finger knuckle print and nail regions are determined. Then, the texture features of finger knuckle are described using four directional patterns based on statistical thresholding. Moreover, the features of nails are described using discrete wavelet transform. In addition, utilizing a cost function in the evolutionary particle swarm optimization algorithm, a set of significance coefficients is applied to the textural features of the hand, and as a result, these features are optimized. Finally, using the d1 distance measurement criterion, the similarity of the images has been calculated and the identity of individuals has been identified with the recognition rate of 92.13%. The experimental results on a set of images collected by the authors, called FKP_Nail, show that the proposed method is more accurate and faster than known existing methods while it uses simpler calculations.

Early diagnosis and treatment of glaucoma can prevent the progression of this disease and the sudden loss of vision. Glaucoma affects the optic disc and optic cup located inside the optic disc. In this paper, first, the optic disc is localized and then segmentation of optic disc and cup is performed to diagnose based on the optic cup to disc ratio (CDR). A Faster Region-based Convolutional Neural Network (Faster-RCNN) with the pre-trained ResNet50 network is used for the optic disc localization step. The segmentation step is performed by the modified U-Net architecture using the SE-ResNet50 network as its encoding layers, and finally CDR is evaluated. The Drishti-GS1 and RIM-ONE v3 databases are used to train and test the proposed method and the MESSIDOR database is only used in the test phase. In addition, for segmentation of optic disc and cup, two approaches are proposed to consider the optic disc and cup annotations in the Drishti-GS1 data set’s ground truth. In the second proposed approach and according to the F1-score criteria, the result of optic cup and disc segmentation for Drishti-GS1 data set is 0.93 and 0.97, respectively, for RIM-ONE v3 data set is 0.79 and 0.95, respectively, and for MESSIDOR data set is 0.84 and 0.93, respectively, which is competitive with other works.

Automatic detection of the angle and position of saffron flowers is an essential step for automatic processing of saffron flowers. Due to the diversity of saffron flowers, traditional algorithms can not yield acceptable accuracy. In this paper, an algorithm based on the deep convolutional networks is proposed. In the proposed architecture, the image is divided into a number of small square areas, and in each region, the existence of the flower is estimated. Also, the relative location suitable for cutting the saffron flower is estimated relative to the center of the region. To estimate the angle, the 360-degree range is first divided into several parts and the angle of the flower is classified to one of the values. Then, the relative angle with the center of the region is estimated. In order to evaluate the performance of the proposed algorithm, a dataset of 163 images and 3035 flowers has been collected, and the parameters for each flower are annotated by an expert. The evaluation of the proposed algorithm shows that more than 95% of the flowers are correctly estimated.

Image super-resolution is a classic image processing issue that aims to create high-resolution images from low-resolution images. Although many algorithms in this field have been proposed so far, effective quality evaluationof such images remains a challenging research area. Conventional image quality assessment measures are not sufficiently consistent with subjective judgments on these images. Therefore, it is very important to provide specific quality assessment methods for image super-resolution. In this paper, we propose a no-reference quality evaluation method for super-resolved images that, by learning a dictionary on high-resolution images and representing super-resolved blocks, produce local features that can describe super-resolution degradations well. These features are pooled togetherby a suitable pyramidal approach and produce a global feature vector of the image. These vectors and subjective quality scores are ultimately used to train a regression model. Experimental results show that this method is not only simple and high speed, but also does not require large volumes of training data and is more efficient than existing methods.

Tracking multiple objects (MOT) is an important topic in almost all computer vision-related areas. One of the most vital challenges in front of MOT’s algorithms is data association, particularly when partial or complete occlusions occur. On the other hand, the human can handle this challenge in everyday scenarios for example while driving a car on a highway or riding a bicycle. Accordingly, we used a brain-inspired method to propose an MOT algorithm that can overcome the above challenge. The proposed method is based on dynamic neural field as a brain-inspired algorithm that can mimic both neural and cognitive functions of the brain. Besides, we benefited from computer vision techniques to find targets and extract features such as their locations, directions, and velocities. We applied our method on an online dataset containing videos recorded from natural movements of zebrafish larvae. Evaluation results using two metrics MOTA and MOTP showed that the proposed method has a promising performance compared to the state-of-the-art algorithms. It can associate all information correctly both in the presence and absence of occlusion events.