1.

Introduction

The main goal of this special section was to have an overview of recent findings in theory, methodology, and applications in the field of image and video analysis, detection, and recognition. In addition to selected contributions from the Proceedings of the 2017 International Conference on Image Analysis and Processing (ICIAP 2017), a well-established conference organized biennially by the Italian Member Society (CVPL) of the International Association for Pattern Recognition (IAPR), contributions were received from all around the world. After a rigorous, multiple-round peer-review process, 29 papers were accepted for publication.

Not surprisingly, the contributions in this special section show that image segmentation and feature extraction remain the most attractive subjects for many researchers, while the increasing role of machine learning is confirmed when complex tasks typical of image classification and recognition are considered. We present a brief overview of the 29 contributions below.

2.

Contributions

The special section opens with two papers on image filtering. Houet al. https://doi.org/10.1117/1.JEI.27.5.051207 develop an L1-based nonlocal total variational model based on Retinex theory for image restoration. Kennedy https://doi.org/10.1117/1.JEI.27.5.051219 describes a procedure for designing highly isotropic separable filters with steerable responses, vanishing moments, and configurable scale.

The next four papers focus on traditional tasks, such as character representation and recognition, with approaches based on classic as well as modern machine learning methods. In the work of ChenChao https://doi.org/10.1117/1.JEI.27.5.051202, skeletonization is used to represent Chinese calligraphy characters. Obaidullahet al. https://doi.org/10.1117/1.JEI.27.5.051214 employ Extreme Learning Machines for handwritten Indic script identification. The Hindi handwritten character recognition problem is addressed by YadavPurwar https://doi.org/10.1117/1.JEI.27.5.051216 using supervised classifiers, namely support vector machines and multilayer perceptrons. Handwriting analysis is the subject of the work of Guarneraet al. https://doi.org/10.1117/1.JEI.27.5.051230, who present GRAPHJ, a tool designed to implement a real forensic protocol adopted in Italy.

The special section continues with six papers on image segmentation. OhYou https://doi.org/10.1117/1.JEI.27.5.051204 investigate a visual saliency region detection method based on soft- and hard-segmentation, while Zhanget al. https://doi.org/10.1117/1.JEI.27.5.051227 propose a graph-based approach inspired by multiple Gestalt principles to detect salient regions. Gaoet al. https://doi.org/10.1117/1.JEI.27.5.051211 propose an active contour method for image segmentation combining color, texture, and saliency. Kothapalliet al. https://doi.org/10.1117/1.JEI.27.5.051226 combine fractional derivatives with information sets to improve the performance of traditional edge detectors. In recent years, deep convolutional neural networks have shown state-of-the-art performance in many image segmentation tasks. Wanget al. https://doi.org/10.1117/1.JEI.27.5.051223 discuss contextual aspects of deep convolutional neural networks for semantic image segmentation, while Gabrielet al. https://doi.org/10.1117/1.JEI.27.5.051228 analyze the discriminative generalized Hough transform in a deep CNN context for pedestrian and car detection.

Two papers deal with the subject of medical image analysis. ÖztürkAkdemir https://doi.org/10.1117/1.JEI.27.5.051213 propose a rescaling method for histopathological images, while Kumariet al. https://doi.org/10.1117/1.JEI.27.5.051208 present a method to develop anatomically realistic breast phantoms using several dispersion models.

Six papers focus on feature extraction. Jiet al. https://doi.org/10.1117/1.JEI.27.5.051205 propose an algorithm that analyzes the state of the eye and mouth, an important step in fatigue detection, by extracting contour features. MisraLaskar https://doi.org/10.1117/1.JEI.27.5.051209 develop a practical hand detection system using sparse texture features and color–texture features. SaeedKhan https://doi.org/10.1117/1.JEI.27.5.051220 address the problem of unconstrained ear recognition by combining traditional ear recognition and ear-based soft biometric traits. PetrelliDi Stefano https://doi.org/10.1117/1.JEI.27.5.051212 use color and depth information for content-based image retrieval of RGB-D imagery. Zhuoet al. https://doi.org/10.1117/1.JEI.27.5.051222 describe a lightweight convolutional network model for vehicle classification. Zwemeret al. https://doi.org/10.1117/1.JEI.27.5.051225 propose a learning-based vehicle make and model recognition system.

Three papers explore various machine learning applications in image processing and analysis. WatsonBhalerao https://doi.org/10.1117/1.JEI.27.5.051215 explore person reidentification by means of deep foreground appearance models. Caiet al. https://doi.org/10.1117/1.JEI.27.5.051218 use depth motion maps and improved convolutional neural networks for human action recognition. Biancoet al. https://doi.org/10.1117/1.JEI.27.5.051221 design a multiscale fully convolutional network architecture for image saliency estimation.

The remaining six papers are devoted to video processing and analysis. Zhuoet al. https://doi.org/10.1117/1.JEI.27.5.051203 deal with vehicle color recognition in surveillance videos using a hierarchical fine-tuning strategy. Meurieet al. https://doi.org/10.1117/1.JEI.27.5.051210 present a human detection method based on a genetically optimized ensemble of classifiers. Thirumurthyet al. https://doi.org/10.1117/1.JEI.27.5.051217 propose statistical-based approaches for change detection in scene regions. Spampinatoet al. https://doi.org/10.1117/1.JEI.27.5.051224 present a low-cost rototranslational video stabilization algorithm. RamadanTairi https://doi.org/10.1117/1.JEI.27.5.051206 investigate salient video object segmentation using boundary connectedness and space-time salient regions. Finally, Bornemanet al. https://doi.org/10.1117/1.JEI.27.5.051229 develop a technique to characterize information content in observed motion using fractal complexity analysis of optical flow.