Owing to the rapid advancements in smartphone technology, there is an emerging need for a technology that can detect banknotes and coins to assist visually impaired people using the cameras embedded in smartphones. Previous studies have mostly used handcrafted feature-based methods, such as scale-invariant feature transform or speeded-up robust features, which cannot produce robust detection results for banknotes or coins captured in various backgrounds and environments. With the recent advancement in deep learning technology, some studies have been conducted on banknote and coin detection using a deep convolutional neural network (CNN). However, these studies also showed degraded performance depending on the changes in background and environment. To overcome these drawbacks, this paper proposes a three-stage detection technology for new banknotes and coins by applying faster region-based CNN, geometric constraints, and the residual network (ResNet). In the experiment performed using the open database of Jordanian dinar (JOD) and 6,400 images of eight types of Korean won banknotes and coins obtained using our smartphones, the proposed method exhibited a better detection performance than the state-of-the-art methods based on handcrafted features and deep features.

Among the existing biometrics methods, finger-vein recognition is beneficial because finger-veins patterns are located under the skin and thus difficult to forge. Moreover, user convenience is high because non-invasive image capturing devices are used for recognition. In real environments, however, optical blur can occur while capturing finger-vein images due to both skin scattering blur caused by light scattering in the skin layer and lens focus mismatch caused by finger movement. The blurred images generated in this manner can cause severe performance degradation for finger-vein recognition. The majority of the previous studies addressed the restoration method of skin scattering blurred images; however, only limited studies have addressed the restoration of optically blurred images. Even the previous studies on the restoration of optical blur restoration have performed restoration based on the estimation of the accurate point spread function (PSF) for a specific image-capturing device. Thus, it is difficult to apply these methods to finger-vein images acquired by different devices. To address this problem, this paper proposes a new method for restoring optically blurred finger-vein images using a modified conditional generative adversarial network (conditional GAN) and recognizing the restored finger-vein images using a deep convolutional neural network (CNN). The results of the experiment performed using two open databases, the Shandong University homologous multimodal traits (SDUMLA-HMT) finger-vein database and Hong Kong Polytechnic University finger-image database (version 1) confirmed that the proposed method outperforms the existing methods.

In recent years, the importance of semantic segmentation has been widely recognized and the field has been actively studied. The existing state-of-the-art segmentation methods show high performance for bright and clear images. However, in low light or nighttime environments, images are blurred and noise increases due to the nature of the camera sensor, which makes it very difficult to perform segmentation for various objects. For this reason, there are few previous studies on multi-class segmentation in low light or nighttime environments. To address this challenge, we propose a modified cycle generative adversarial network (CycleGAN)-based multi-class segmentation method that improves multi-class segmentation performance for low light images. In this study, we used low light databases generated by two road scene open databases that provide segmentation labels, which are the Cambridge-driving labeled video database (CamVid) and Karlsruhe Institute of Technology and Toyota Technological Institute at Chicago (KITTI) database. Consequently, the proposed method showed superior segmentation performance compared with the other state-of-the-art methods.

In the current field of human recognition, most of the research being performed currently is focused on re-identification of different body images taken by several cameras in an outdoor environment. On the other hand, there is almost no research being performed on indoor human recognition. Previous research on indoor recognition has mainly focused on face recognition because the camera is usually closer to a person in an indoor environment than an outdoor environment. However, due to the nature of indoor surveillance cameras, which are installed near the ceiling and capture images from above in a downward direction, people do not look directly at the cameras in most cases. Thus, it is often difficult to capture front face images, and when this is the case, facial recognition accuracy is greatly reduced. To overcome this problem, we can consider using the face and body for human recognition. However, when images are captured by indoor cameras rather than outdoor cameras, in many cases only part of the target body is included in the camera viewing angle and only part of the body is captured, which reduces the accuracy of human recognition. To address all of these problems, this paper proposes a multimodal human recognition method that uses both the face and body and is based on a deep convolutional neural network (CNN). Specifically, to solve the problem of not capturing part of the body, the results of recognizing the face and body through separate CNNs of VGG Face-16 and ResNet-50 are combined based on the score-level fusion by Weighted Sum rule to improve recognition performance. The results of experiments conducted using the custom-made Dongguk face and body database (DFB-DB1) and the open ChokePoint database demonstrate that the method proposed in this study achieves high recognition accuracy (the equal error rates of 1.52% and 0.58%, respectively) in comparison to face or body single modality-based recognition and other methods used in previous studies.

In recent years, the importance of the semantic segmentation field has been increasingly emphasized because autonomous vehicle and artificial intelligence (AI)-based robot technology are being researched extensively; and methods for accurately recognizing objects are required. Previous state-of-the-art segmentation methods have been proven to be effective for databases obtained during daytime. However, in extremely low light or nighttime environments, the shape and color information of objects are very small or disappear due to an insufficient amount of external light, which makes it difficult to train the segmentation network and significantly degrades performance. In our previous work, segmentation performance in a low light environment was improved using the enhancement-based segmentation method. However, low light images could not be restored precisely and segmentation performance improvement was limited because only per-pixel loss functions were used when training the enhancement network. To overcome these drawbacks, we propose a low light image segmentation method based on a modified perceptual cycle generative adversarial network (CycleGAN). Perceptual image enhancement was performed using our network, which significantly improved segmentation performance. Unlike the existing perceptual loss, the Euclidean distance of the feature maps extracted from the pretrained segmentation network was used. In our experiments, we used low light databases generated from two famous road scene open databases, which are Cambridge-driving Labeled Video Database (CamVid) and Karlsruhe Institute of Technology and Toyota Technological Institute at Chicago (KITTI), and confirmed that our proposed method shows better segmentation performance in extremely low light environments than the existing state-of-the art methods.

Stem removal from harvested fruits remains one of the most labor-intensive tasks in fruit harvesting, directly affecting the fruit quality and marketability. Accurate and rapid fruit and stem segmentation techniques are essential for automating this process. This study proposes an enhanced You Only Look Once (YOLO) model, AppleStem (AS)-YOLO, which uses a ghost bottleneck and global attention mechanism to segment apple stems. The proposed model reduces the number of parameters and enhances the computational efficiency using the ghost bottleneck while improving feature extraction capabilities using the global attention mechanism. The model was evaluated using both a custom-built and an open dataset, which will be later released to ensure reproducibility. Experimental results demonstrated that the AS-YOLO model achieved high accuracy, with a mean average precision (mAP)@50 of 0.956 and mAP@50–95 of 0.782 across all classes, along with a real-time inference speed of 129.8 frames per second (FPS). Compared with state-of-the-art segmentation models, AS-YOLO exhibited superior performance. The proposed AS-YOLO model demonstrates the potential for real-time application in automated fruit-harvesting systems, contributing to the advancement of agricultural automation.