Multimedia Tools and Applications

In this paper, a novel vision assisted genetic algorithm based navigational controller has been designed for smooth and collision-free path generation of a humanoid robot. Here, sensory information regarding the nearest obstacle distance and path left to the destination are considered as the inputs to the genetic algorithm controller, and necessary turning angle is generated as the required output to avoid the obstacles present in the path and advance towards the destination. The vision based technique is integrated along with the sensor based navigational model to assist in deciding a safe direction of turn in case the humanoid encounters a dead end situation while negotiating with complicated obstacle settings. The developed model has been verified by navigational analysis of a NAO humanoid in a V-REP simulation arena. The simulation results are also validated against an experimental set-up prepared under laboratory conditions that resembles the simulation arena. The results obtained from both the platforms are compared in terms of selected navigational parameters, and a close agreement has been found between them with a minimal percentage of errors. Finally, the developed model is also evaluated against other existing navigational schemes, and substantial performance improvements have been observed.

The 3-factor authentication control system is aim at providing maximum security in company’s stores, special utilities and premises where valuables are stored. A developed security system with automatic sensing was introduced using the integration of Radio Frequency Identification (RFID) card tagging system, fingerprint sensing biometric security system and Global System for Mobile communication (GSM) to send token generated by the system to the authorized user’s access cell-phone. The accuracy of the system was measured virtually in a simulation screen. Results and performance evaluation test shows over 98% accuracy in access granted to access denial, for both right and wrong RFID card and correct to incorrect fingerprint scanning respectively. This consequently, shows satisfaction in the performance of the system test, which proves the validity and efficiency of the system by ensuring full integrity of the door lock in any premises as the case may be.

Video person re-identification (video Re-ID) is a key technology applied to video surveillance and security. Typical person re-identification is designed to retrieve the correct match of the target image (query) from gallery images, while video Re-ID extends this to query from gallery videos. The main factors affecting the video Re-ID model are: (i) a high-quality frame-level feature extractor, and (ii) temporal modeling that combines frame-level features into a feature for retrieval. In this work, we use ShuffleNet V2-based lightweight algorithm for video Re-ID, which can meet the demand for practical application and solve the problem of high consumption for computing resources, and maintain high performance. At the same time, the lightweight space attention mechanism Spatial Group-wise Enhance (SGE) module is used to view the person in more detail, which makes the feature representation more compact and effectively improves the retrieval accuracy. Finally, we design an Online Difference Discrimination (ODD) module to measure the feature gap between video frames, and use this module to make different temporal modeling for different quality video sequences. Experiments on three datasets (i.e., iLIDS-VID, PRID2011 and MARS) show that our method is competitive with state-of-the-art methods.

Static video summarization is recognized as an effective way for users to quickly browse and comprehend large numbers of videos. In this paper, we formulate static video summarization as a clustering problem. Inspired by the idea from high density peaks search clustering algorithm, we propose an effective clustering algorithm by integrating important properties of video to gather similar frames into clusters. Finally, all clusters’ center will be collected as static video summarization. Compared with existing clustering-based video summarization approaches, our work can detect frames which are highly relevant and generate representative clusters automatically. We evaluate our proposed work by comparing it with several state-of-the-art clustering-based video summarization methods and some classical clustering algorithms. The experimental results evidence that our proposed method has better performance and efficiency.

This paper presents the system architecture of a collaborative virtual environment in which distributed multidisciplinary teams involved in space exploration activities come together and explore areas of scientific interest of a planet for future missions. The aim is to reduce the current challenges of distributed scientific and engineering meetings that prevent the exploitation of their collaborative potential, as, at present, expertise, tools and datasets are fragmented. This paper investigates the functional characteristics of a software framework that addresses these challenges following the design science research methodology in the context of the space industry and research. An implementation of the proposed architecture and a validation process with end users, based on the execution of different use cases, are described. These use cases cover relevant aspects of real science analysis and operation, including planetary data visualization, as the system aims at being used in future European missions. This validation suggests that the system has the potential to enhance the way space scientists will conduct space science research in the future.

Dermoscopy is a noninvasive way to examine and diagnose skin lesions, e.g. nevus and melanoma, and is a critical step for skin cancer detection. Accurate classification of dermoscopic images can detect skin cancer at an early stage and bring social and economic impact to patients and communities. Using deep learning methods to classify dermoscopic images has shown superior performance, but existing research often overlooks the class imbalance in the data. In addition, although a handful of public datasets are available for skin cancer research, these datasets are generally not large enough for deep learning algorithms to produce accurate results. In this paper, we propose to use data augmentation and generative adversarial networks (GAN) to tackle class-imbalanced dermoscopic image classification. Our main objectives are to determine (1) how state-of-the-art fine-tuned deep learning models perform on class-imbalanced dermoscopic images, (2) whether data augmentation and GAN can help alleviate class imbalances to improve classification accuracy, and (3) which method is more effective in addressing the class imbalance. By using public datasets and a carefully designed framework to generate augmented images and synthetic images, our research provides clear answers to these questions. Code and data used in the study are available at: https://github.com/mjan2021/Dermoscopic-image-classification.git

As an important part of multimedia analysis applications, feature selection has attracted much attention during the past decades. Lots of feature selection methods have been proposed, but most of them neglect to consider the correlation between the selected features, which leads to the feature redundancy problem. In this paper, we propose a novel supervised feature selection method, termed as Uncorrelated Discriminant Sparse Regression (UDSR). This method is an organic combination of discriminant sparse regression and uncorrelated constraint. In this method, the discriminant sparse regression ensures the discriminant power of the selected features, and the uncorrelated constraint avoids the redundancy of selected features. Thus the features selected by our method are not only discriminative but also uncorrelated with each other. The method can be applied to a wide range of multimedia applications. Experiments are conducted on two video datasets and four image datasets. The experimental results show that the proposed method has better performance for multimedia analysis, compared to the baseline and six state-of-the-art relative methods.

Aiming at the lack of coarse-grained features in butterfly image classification and recognition research, low recognition accuracy, and limited spatial invariance to input data, this paper proposes a butterfly image classification method based on U-Net and STL-MSDNet. Firstly, in order to reduce the influence of complex backgrounds on butterfly image recognition, the U-Net model is used to segment the butterfly ecological image. Then, an STL-MSDNet model is proposed to classify butterfly images. In STL-MSDNet, Spatial Transformer Network (STN) is added to reverse the spatial transformation of butterfly images to eliminate the deformation of image butterflies and make the recognition of the classification network simpler and more efficient. Then, the Laplace pyramid (LP) was introduced to replace gaussian down-sampling in MSDNet, and the butterfly images were decomposed into different spatial frequency bands to obtain butterfly feature maps of three scales, and then they were fused to improve the feature extraction capability of the network. The experimental results show that the butterfly image semantic segmentation algorithm based on U-Net has a good effect and is suitable for the field of image segmentation in complex backgrounds. Compared with MSDNet, DenseNet and traditional classification algorithms, the butterfly image classification model based on STL-MSDNet proposed in this paper has a better effect, better robustness, and a higher recognition rate. The method proposed in this paper solved the problem of low accuracy of classification of butterfly images in complex backgrounds by existing methods, and obtains a classification accuracy of 93.8%, indicating that it has good results in the fine classification of butterflies and can be applied to butterfly identification and realize the application of butterfly ecological research.

In this paper, we analyze a prefetching mechanism for image content in social networking services (SNS) based on content writers that attract a user’s interest. Our prefetching scheme aims to shorten transmission delays and enhance the accessibility to big size images included in SNS content items. The prefetching scheme deals with users of interest instead of the content of interest. Whenever a user downloads new SNS content, his/her client then prefetches the original high-resolution images uploaded by writers of interest to the user. The performance simulations show that the prefetching scheme is well suited if the user’s access pattern is highly skewed to a few number of content writers, and that the prefetching scheme gives users high-quality images without significant increase of access delay.