Special Issue on Intelligent Processing on Image and Optical Information II

1. Introduction

Intelligent image and optical information processing have paved the way for the recent epoch of new intelligence and information. Certainly, information acquired by various imaging techniques is of tremendous value; thus, an intelligent analysis of them is necessary in order to make the best use of it.

The objectives of intelligent processing range from the refinement of raw data to the symbolic representation and visualization of the real world. The image and optical information are acquired by various sources and imaging methods [1]. Thus, the extraction and manipulation of the descriptive features are essential for such a task [2]. It comes through unsupervised or supervised learning based on statistical and mathematical models or computational algorithms [3,4]. With recent advances in computing power and learning algorithms, many applications have become more practical and further development is expected [5].

This Special Issue focuses on the intelligent processing of images and optical information such as object detection and classification and three-dimensional imaging. With recent advances in deep learning, detection and classification of images are often performed via convolutional neural networks. Depth information is essential for displaying and investigating the three-dimensional structure of the real world. Stereo-matching and digital holography provide depth; 3D scanning technique captures the geometry and shape of objects in three-dimensional space. In addition to them, image segmentation and enhancement are presented as key steps for image analysis and raw data refinement. Several studies on image processing applications in medical diagnosis, manufacturing inspection, and target tracking demonstrate the significant value and potential benefits of intelligent processing in various domains and industries.

A total of 13 papers were verified through a thorough review process. Many valuable and up-to-most recent technologies are provided to solve the real problems in selected papers. The second volume of the Special Issue on the topic is closed; more in-depth research of the same topic is expected in the third volume of the Special Issue. It is anticipated that the scope of intelligent processing would be even broader in the future.

2. Intelligent Processing on Image and Optical Information Vol. II

This Special Issue was introduced to collect the latest research on relevant topics, and more importantly, to address the current practical and theoretical challenges. In the following, the papers are categorized into several subtopics: detection and classification, three-dimensional imaging, image segmentation and enhancement, and image processing applications for medical diagnosis, manufacturing inspection, and target tracking.

2.1. Detection and Classification

Image detection and classification with deep learning have become important topics in intelligent image processing. In the first paper of this category, entitled ‘Compact and Accurate Scene Text Detector’, Minjun Jeon and Young-Seob Jeong [6] proposed scene text detection with a new efficient deep learning model. The authors designed a balanced decoder to detect word boxes in given images, which is much faster than the existing ones.

Action recognition is an important topic in video analysis. It deals with temporal dynamics in a sequence of frames. A paper, entitled ‘Fine-Grained Action Recognition’ by Fang Liu, Liang Zhao, Xiaochun Cheng, Qin Dai, and Xiangbin Shi, and Jianzhong Qiao [7], proposes an action recognition model using a graph structure to describe relationships between the mid-level patches.

Ji-Hyeon Yoo, Ho-Jin Jung, Yi-Sue Jung, Yoon-Bee Kim, Chang-Jae Lee, Sung-Tae Shin, and Han-Ul Yoon [8], in their paper on ‘Classifying Upper Arm Gym-Workouts via Convolutional Neural Network by Imputing a Biopotential-Kinematic Relationship’, proposed a systemic approach to upper arm gym-workout classification according to spatio-temporal features depicted by biopotential as well as joint kinematics.

As the deep-learning model develops, optimization becomes increasingly crucial in the training process. Dokkyun Yi, Sangmin Ji, and Jieun Park [9] described in their article, ‘An Adaptive Optimization Method Based on Learning Rate Schedule for Neural Networks’ a novel method solving the non-stopping problem of the momentum-based optimization.

2.2. Three-Dimensional Imaging

Advances in three-dimensional imaging have enabled the capture and reconstruction of three-dimensional information. The paper entitled ‘Improved Cost Computation and Adaptive Shape Guided Filter for Local Stereo Matching of Low Texture Stereo Images’ by Hua Liu, Rui Wang, Yuanping Xia, and Xiaoming Zhang [10] proposes an efficient and effective matching cost measurement and an adaptive shape guided filter-based matching cost aggregation method to improve the stereo matching performance for large textureless regions.

Ketao Yan, Lin Chang, Michalis Andrianakis, Vivi Tornari, and Yingjie Yu [11], in their paper on ‘Deep Learning-Based Wrapped Phase Denoising Method for Application in Digital Holographic Speckle Pattern Interferometry’, introduced a deep-learning based method for denoising interferograms obtained by digital holographic speckle pattern interferometry. The proposed denoising method can serve in the structural diagnosis of artworks by accurately detecting defects in complex defect topography maps.

The structured-light technique is an effective method for indoor three-dimensional measurements. In the paper entitled ‘3D Snow Sculpture Reconstruction Based on Structured-Light 3D Vision Measurement’, Wancun Liu, Liguo Zhang, Xiaolin Zhang, and Lianfu Han [12] proposed a 3D vision measurement method to improve resistance to noise of measuring systems, which ensures normal operation of a structured-light sensor in the wild without changing its components, and the method is applied in 3D reconstruction of snow sculpture.

2.3. Inage Segmentation and Enhancement

Multi-level thresholding is a direct and effective method for image segmentation. The paper entitled ‘Renyi’s Entropy Based Multilevel Thresholding Using a Novel Meta-Heuristics Algorithm’ by Wei Liu, Yongkun Huang, Zhiwei Ye, Wencheng Cai, Shuai Yang, Xiaochun Cheng, and Ibrahim Frank [13] proposes a meta-heuristics algorithm based on the breeding mechanism of Chinese hybrid rice in order to seek the optimal multi-level thresholds.

Thermal imaging converts the temperature distribution emitted by objects into a visible image. In the paper entitled ‘Multi-Scale Ensemble Learning for Thermal Image Enhancement’, Yuseok Ban, and Kyungjae Lee [14] proposed a multi-scale ensemble learning method in different image scale conditions, which has a novel parallel architecture leveraging the confidence maps of multiple scales.

2.4. Image Processing Applications

Several practical solutions are emphasized in this subsection. The automatic estimation of tortuosity is essential in medical diagnosis. In the paper entitled Automatic Tortuosity Estimation of Nerve Fibers and Retinal Vessels in Ophthalmic Images’, Honghan Chen, Bang Chen, Dan Zhang, Jiong Zhang, and Jiang Liu, and Yitian Zhao [15] proposed an automated framework for tortuosity estimation for corneal nerves and retinal vessels.

The paper entitled ‘The Influence of Image Processing and Layer-to-Background Contrast on the Reliability of Flatbed Scanner-Based Characterisation of Additively Manufactured Layer Contours’ by David Blanco, Pedro Fernández, Alejandro Fernández, Braulio J. Alvarez, and José Carlos Rico [16] evaluates the influence of image processing and layer-to-background contrast characteristics upon contour reconstruction quality under a metrological perspective.

Strain gauges are commonly used for tension tests to obtain the strain of metal t specimens. In the paper entitled ‘A Study on Tensile Strain Distribution and Fracture Coordinate of Nanofiber Mat by Digital Image Correlation System’, Nak Gyu Park, Kyung Min Hong, and Kyu Hyeung Kwon [17] employed the digital image correlation method to measure displacement and calculate strain for all areas of the specimen.

Drones have played significant roles in security and surveillance. The paper entitled ‘Moving Vehicle Tracking with a Moving Drone Based on Track Association’ by Seokwon Yeom and Don-Ho Nam [18] proposes a new data association scheme for target tracking comprising the hypothesis testing and track fusion.