SỬ DỤNG PHƯƠNG PHÁP SUPPORT VECTOR MACHINE VÀ CONVOLUTIONAL NEURAL NETWORK ĐỂ PHÂN LOẠI CẢM XÚC CỦA KHUÔN MẶT

Phân loại cảm xúc khuôn mặt thông qua việc nhận diện nét mặt hiện là một trong những bài toán được nhiều nhà nghiên cứu quan tâm. Với mục đích hỗ trợ được người dùng trong việc nhận diện được cảm xúc khuôn mặt để sử dụng nghiên cứu trong các lĩnh vực như khoa học lâm sàng hay khoa học hành vi. Thách thức với bài toán này là nét mặt của con người có sự tương đồng, trùng lặp trong các biểu thị cảm xúc khác nhau. Trong bài báo này, nhóm tác giả đề xuất sử dụng phương pháp Support Vector Machine – SVM kết hợp với mạng thần kinh chuyển đổi để phân loại cảm xúc trên khuôn mặt trên bộ dữ liệu FER và xây dựng 3 mô hình chiến lược để tiến hành các thí nghiệm. Việc xác định cảm xúc chính xác của khuôn mặt luôn khó khăn. Các kết quả của thực nghiệm đã cho thấy rằng mô hình Convolutional Neural Network - CNN chính xác hơn khi so sánh với mô hình Histogram Of Orientation Gradient + Support Vector Machine - HOG +SVM . Mô hình CNN lấy hình ảnh thực làm đầu vào và mô hình CNN nhập hình ảnh kết hợp có kết quả gần đúng với nhau và có tính chất ổn định hơn.

Face Verification Across Age Progression using Enhanced Convolution Neural Network

This paper proposes a deep learning method for facial verification of aging subjects. Facial aging is a texture and shape variations that affect the human face as time progresses. Accordingly, there is a demand to develop robust methods to verify facial images when they age. In this paper, a deep learning method based on GoogLeNet pre-trained convolution network fused with Histogram Orientation Gradient (HOG) and Local Binary Pattern (LBP) feature descriptors have been applied for feature extraction and classification. The experiments are based on the facial images collected from MORPH and FG-Net benchmarked datasets. Euclidean distance has been used to measure the similarity between pairs of feature vectors with the age gap. Experiments results show an improvement in the validation accuracy conducted on the FG-NET database, which it reached 100%, while with MORPH database the validation accuracy is 99.8%. The proposed method has better performance and higher accuracy than current state-of-the-art methods.

A Non-Classical Analytical Approach for Vibration Analysis of Isotropic and Fgm Plate Containing a Star Shaped Crack

ABSTRACTA non-classical analytical model for vibration analysis of thin isotropic and FGM plate containing multiple part-through cracks (star shaped) of arbitrary orientation is proposed. A plate containing four concentric cracks of arbitrary orientation in the form of continuous line is considered for analysis. The proposed governing equation is derived based on classical plate theory and modified couple stress theory. Line spring model is modified to accommodate all the crack terms. The application of Berger’s formulation introduces nonlinearities in the governing equation and then the Galerkin’s method is applied for solving final governing equation. Results for fundamental frequencies for different values of crack length, crack orientation, gradient index and material length scale parameters are presented for two different boundary conditions. Furthermore, to study the phenomenon of bending hardening/softening in a cracked plate, the frequency response curves are plotted for the parameters stated above. Based on the outcomes of this study, it can be concluded that stiffness of the plate is severely affected by the presence of multiple cracks and the stiffness goes on decreasing with increase in number of cracks thereby affecting the fundamental frequency.

A Novel Scale Insensitive KCF Tracker Based on HOG and Color Features

Kernelized Correlation Filters (KCF) for visual tracking have received much attention due to their fast speed and outstanding performances in real scenarios. However, the KCF sometimes still fails to track the targets with different scales, and it may drift because the target response is fixed and the original histogram of orientation gradient (HOG) features cannot represent the targets well. In this paper, we propose a novel fast tracker, which is based on KCF and insensitive to scale changes by learning two independent correlation filters (CFs) where one filter is designed for position estimation and the other is for scale estimation. In addition, it can adaptively change the target response and multiple features are integrated to improve the performance for our tracker. Finally, we employ an adaptive high confidence filters updating scheme to avoid errors. Evaluated on the popular OTB50 and OTB100 datasets, our proposed trackers show superior performances in terms of efficiency and accuracy compared to the existing methods.

Estimation of geometrically necessary dislocation density from filtered EBSD data by a local linear adaptation of smoothing splines

An implementation of smoothing splines is proposed to reduce orientation noise in electron backscatter diffraction (EBSD) data, and subsequently estimate more accurate geometrically necessary dislocation (GND) densities. The local linear adaptation of smoothing splines (LLASS) filter has two advantages over classical implementations of smoothing splines: (1) it allows for an intuitive calibration of the fitting versus smoothing trade-off and (2) it can be applied directly and in the same manner to both square and hexagonal grids, and to 2D as well as to 3D EBSD data sets. Furthermore, the LLASS filter calculates the filtered orientation gradient, which is actually at the core of the method and which is subsequently used to calculate the GND density. The LLASS filter is applied on a simulated low-misorientation-angle boundary corrupted by artificial orientation noise (on a square grid), and on experimental EBSD data of a compressed Ni-base superalloy (acquired on a square grid) and of a dual austenitic/martensitic steel (acquired on an hexagonal grid). The LLASS filter leads to lower GND density values as compared to raw EBSD data sets, as a result of orientation noise being reduced, while preserving true GND structures. In addition, the results are compared with those of filters available in theMTEXtoolbox.