Brazil

The study of shape objects, in special, its description and recognition is a field of intense research in computer vision. This paper presents a novel approach for shape descriptor based on two methods: the randomized neural network (RNN), which has a single hidden neuron layer and a training algorithm much faster than the traditional \emph{backpropagation} approach, and the Fourier transform, a traditional signal analysis tool that enables us to decompose a signal into its frequencies. By combining these methods, we are able to compute a robust shape signature, which is capable of recognizing very different types of images, from occluded shapes to large sets of images under affine transformations. We compared the proposed approach with several shape analysis methods. The results indicate that our proposed approach can be successfully applied in different shape analysis problems.

IWSSIP 2020

Shape Classification Using Randomized Neural Network Descriptors Extracted from the Fourier Spectrum

fourier spectrum.

randomized neural network

shape recognition

shape analysis

The purpose of the AAAI conference is to promote research in artificial intelligence (AI) and scientific exchange among AI researchers, practitioners, scientists, and engineers in affiliated disciplines. 

AAAI 2021

technical paper

The 27th International Conference on Systems, Signals and Image Processing.

Transmission Electron Microscopy (TEM) is one of the most important tools for virus detection and identification. It enables researchers to study the characteristics of a virus in order to classify it into groups and families. From morphological characteristics of viruses in TEM images, this paper proposes a computer vision approach for virus classification. For this purpose, we compared ten texture descriptors methods and fused them in order to obtain a higher discriminative signature. We compared this signature to other works that addressed the same virus classification problem. The result achieved by our proposed fusion approach (87.27\%) surpassed the highest accuracies of previous papers, thus proving to be very effective for discriminating virus in TEM images. Thus, we believe that our paper brings a relevant contribution to this important medical problem.

Virus Classification by using a Fusion of Texture Analysis Methods

The human brain is a complex system and many approaches are currently being used to study its disorders. Functional Connectivity is a well-known approach which characterizes the relationship between two or more regions of the brain.
This relationship is inferred from measuring and correlating activity from the brain. On the other hand, Effective Connectivity quantifies activities instead of correlating them. For Functional Connectivity, Phase Locked Value (PLV) and other measures are used. For Effective Connectivity, Transfer Entropy is the main measure used. The problem is: Functional Connectivity describes the flow of information in the complex network of the brain in terms of correlation, and allows time delays computations for
better correlations – but gives no information regarding causality; Transfer Entropy is very accurate regarding causality, but has assessed interactions at only one time delay. This paper will address these issues by analysing a data set of absence seizures
on both connectivity analyses. PLV (Phase Locking Value) and Transfer Entropy (TE) algorithms will be performed on EEG data from 11 subjects diagnosed with Absence Seizures. Afterwards, both networks will be described in terms of each connectivity
type. The main contribution of this paper is that the network specification of both Functional and Effective Connectivity are complementary when characterizing absence seizures from EEG data.

Functional and Effective Connectivity Characterization of Absence Seizures

Due to the positive relationship between the presence of students in classes and their performance, student attendance assessment is considered essential within the classroom environment, even as a tiring and time-consuming task. We proposed a solution for student attendance control using face recognition with deep one-shot learning and evaluated our approach in different conditions and image capturing devices to confirm that such a pipeline may work in a real-world setting. For better results regarding the high number of false negatives that often occur in uncontrolled environments, we also proposed a face detection stage using HOG and a CNN with Max-Margin Object Detection based features. We achieved accuracy and F1 scores of 97% and 98.4% with an iPhone 7 camera, 91.9% and 94.8% with a Moto G camera, and 51.2% and 61.1% with a WebCam respectively. These experiments reinforce the effectiveness and availability of this approach to the student attendance assessment problem since the recognition pipeline can be either made available for embedded processing with limited computational resources (smartphones), or offered as "Software as a Service" tool.

Automatic Attendance Management System based on Deep One-Shot Learning

Face recognition has been one of the most relevant and explored Biometrics fields. In real-world applications, face recognition methods usually must deal with scenarios where not all probe individuals were seen during training phase (open-set scenarios). Therefore, open-set face recognition is a subject of increasing interest as it deals with identifying individuals in a space where not all faces are known in advance. This is useful in several applications, such as access authentication, on which only a few individuals that have been previously enrolled in a gallery are allowed. The present work introduces a novel approach towards open-set face recognition focusing on small galleries and in enrollment detection, not identity retrieval. A Siamese Network architecture is proposed to learn a model to detect if a face probe is enrolled in the gallery based on a verification-like approach. Promising results were achieved for small galleries, outperforming state-of-the-art methods like HFCN and HPLS, on experiments carried out on Pubfig83, FRGCv1 and LFW datasets. In addition, a new evaluation protocol is introduced for experiments in small galleries.

Open-set Face Recognition for Small Galleries Using Siamese Networks

Automatic speaker recognition is drastically
degraded in presence of noise. This paper focuses on the application of the cubic law and histogram mapping for the textindependent speaker recognition task. Our aim of this study is the application of these two methods in the feature extraction stage of the Power-Normalized Cepstral Coefficients (PNCC) and the conventional Mel Frequency Cepstral Coefficients (MFCC) techniques. Recognition results show that the cubic law combined with the histogram mapping improve the recognition rates.

Cubic Law and MAP Compensation Techniques for Robust Text-Independent Speaker Identification

Dysplasia is a common oral premalignant lesion type that can be classified as mild, moderate and severe. The diagnosis of different types of dysplasia is made by a pathologist through complex and time consuming histological image analysis. The use of computer-aided diagnosis can be applied as a tool to aid and enhance the pathologists decisions. Recently, deep learning based methods are earning more and more attention and have been successfully applied to nuclei segmentation problems in several scenarios. In this paper, we present a method for automated nuclei segmentation on dysplastic oral tissues histological images using convolutional neural networks. We also evaluated the impact of color normalization techniques applied to the automated nuclei segmentation task on hematoxylin-eosin stained histological images. The proposed method achieves the best results overall when validated against other segmentation methods using a dataset composed of mice tongue histopathological images.

Automated Nuclei Segmentation on Dysplastic Oral Tissues Using CNN

This work presents the Link-Quality Aware and Thermal Aware On-Demand Routing (LATOR) protocol, a proposal for routing in Wireless Body Area Networks (WBAN). LATOR improves Packet Delivery Rate (PDR) in intra-WBAN communication with node overheating prevention. This protocol avoids overheating considering only each node’s temperature without exchanging temperature information between neighboring nodes, which allows a decrease in control traffic. LATOR also considers route selection based on Link Quality Information (LQI). Initially the Received Signal Strength Indicator (RSSI) was considered as LQI. The proposal was implemented and evaluated using the Castalia framework over the OMNET++ simulator. Experimental results show that LATOR allows better PDR performance, especially when the direct communication presents low quality and there is an alternative path with better quality. At the same time, overheating in the network is avoided and PDR adequate performance is maintained. Although LATOR increases end-to-end latency, it can be considered a viable alternative for WBANs applications.

LATOR: Link-Quality Aware and Thermal Aware On-Demand Routing Protocol for WBAN

This presentation is composed of two main contributions. The first one concerns the segmentation of plantar foot thermal images. It consists of using the deep learning method U-Net to segment the thermal images. U-Net is trained by combining the two types of images (thermal and color) given by the thermal camera FLIR ONE Pro. Results show that this multimodal approach performs better than the one using only
thermal images, especially for difficult cases. The second part is devoted to a transversal clinical study conducted within the Hospital National Dos de Mayo in Lima, Peru. 122 type II diabetic patients without ulcer were recruited. These individuals were classified into three risk groups of developing a foot ulcer. This
classification is based on a medical examination: a low-risk group (R0), a medium-risk group (R1) and finally a high-risk group (R2).The results demonstrate that thermal camera temperature assessment could help in the diagnosis of diabetic foot.

Segmentation of Plantar Foot Thermal Images: Application to Diabetic Foot Diagnosis

Query performance prediction (QPP) is the namegiven to approaches that aim at predicting, usually post-query, thequality of results generated by an IR system. Yet, the indirect waythat standard QPP approaches estimate performance is opaque,has little relation to the way users evaluate queries, and islimited in application. An alternative, more challenging way tolook at the QPP problem is as arelevance predictionproblem.Instead of resulting in a quality estimate, relevance predictionaims at predicting the relevance of the top-k results in a rank.When accurate, those predictions can be directly employed as amechanism to improve the initial rank. In this work, we explorea learning framework based on sequence labeling combined withcontextual features for relevance prediction. Sequence labelingis a common approach to deal with sequential data, learningfrom the sequence to generate a sequence of predicted labels.The context within a rank, that is, the neighborhood of rankedobjects, has been successfully explored within rank-aggregationand re-ranking scenarios, and here we extend the idea to devisean effective relevance prediction feature. We show that thecombination of both can beat the baseline relevance predictionapproach, performing effective prediction, that is accurate for upto k = 30 positions.

Contextual Features and Sequence Labeling Techniques for Relevance Prediction in Retrieval

Personalized music systems usually rely on manual song annotations (tags) as a mechanism for querying and navigating large music collections. However, the manual annotation is a hard task given the large amount of music available nowadays. Automatic song annotation based on content analysis is a potential solution to this problem and has recently been gaining attention. In this work, we propose to extend the Stacking prediction framework to use Convolutional Neural Networks (CNNs) in order to improve music tag annotation task. In general, the Stacked prediction consists of a technique in which the output of the first stage of learning is used as input in the second stage. In our work we have tried two proposals of extension. The first one consists of using the weights learned by the CNN in the first stage of training as input in the second stage. In the second proposal, we use the autoencoder technique with the weights learned by CNN in the first stage to generate images 50% smaller than the ones used as input of the first stage. The weights and the images obtained in the first stage are used as input in the second stage. We evaluated our proposals with five different CNN models in three datasets well-known in the literature (FMA, MillionSong, and MagnaTagATune), obtaining interesting results.

Premium content

Downloads

Next from IWSSIP 2020

Virus Classification by using a Fusion of Texture Analysis Methods

Similar lecture

Unsupervised 3D Learning for Shape Analysis via Multiresolution Instance Discrimination