Proposed Feature Extraction Algorithm Research Articles

DP-AOP: A novel SVM-based antioxidant proteins identifier

The identification of antioxidant proteins is a challenging yet meaningful task, as they can protect against the damage caused by some free radicals. In addition to time-consuming, laborious, and expensive experimental identification methods, efficient identification of antioxidant proteins through machine learning algorithms has become increasingly common. In recent years, researchers have proposed models for identifying antioxidant proteins; unfortunately, although the accuracy of models is already high, their sensitivity is too low, indicating the possibility of overfitting in the model. Therefore, we developed a new model called DP-AOP for the recognition of antioxidant proteins. We used the SMOTE algorithm to balance the dataset, selected Wei's proposed feature extraction algorithm to obtain 473 dimensional feature vectors, and based on the sorting function in MRMD, scored and ranked each feature to obtain a feature set with contribution values ranging from high to low. To effectively reduce the feature dimension, we combined the dynamic programming idea to make the local eight features the optimal subset. After obtaining the 36 dimensional feature vectors, we finally selected 17 features through experimental analysis. The SVM classification algorithm was used to implement the model through the libsvm tool. The model achieved satisfactory performance, with an accuracy rate of 91.076 %, SN of 96.4 %, SP of 85.8 %, MCC of 82.6 %, and F1 core of 91.5 %. Furthermore, we built a free web server to facilitate researchers' subsequent unfolding studies of antioxidant protein recognition. The website is http://112.124.26.17:8003/#/.

Development of Feature Extraction Technique

A problem of personal verification and identification is an actively growing area of research in pattern recognition, computer vision, and digital image processing. A Face acknowledgment system is largely significant move toward in our life. Face acknowledgment method must be proficient to robotically identify images of face. This system mainly used to recognize a person and make available a protection during various features of our life. It is extremely complicated job for investigator to obtain mainly excellent face acknowledgment velocity in a variety of circumstances and benchmarks. Face recognition is grassland of computer visualization to uses faces to recognize or authenticate a human being. Principal Component Analysis (PCA) is accomplished and used for feature extraction and measurement lessening. The feature extraction is used to reduce the dimension of the face space by transforming it into feature representation. Features may be symbolic, numerical or both. The symbolic feature is color and numerical feature is weight. The combined feature extraction of PCA, LDA and Wavelet are used in proposed feature extraction algorithm for human face recognition system. The structure is tested and achieves high recognition rates. Information regarding individuals was stored in a database.

Feature Extraction and Small-Sample Learning of Dexmedetomidine for Neurosurgery on Postoperative Agitation in Patients with Craniocerebral Injury.

Objective. To observe the controlled effect of dexmedetomidine for neurosurgery and the effect on postoperative cognitive function. The main task of this paper is to use data from a small sample. The proposed feature extraction algorithm based on the bilinear convolutional neurological network (BCNN) is based on a small sample of data. BCNN involves the simultaneous extraction of highly discriminative cross-sectional features from the input image using two parallel subnetworks. By optimizing the algorithm to minimize losses, the two subnetworks can be supervised by each other, improving the performance of the network and obtaining accurate recognition results without spending a lot of time adjusting parameters. The mean arterial pressure (MAP) and heart rate (HR) levels of cerebral oxygen metabolism were compared between the two groups before (T0), after (T1), immediately after (T2), and after intubation (T3). In the observation group, MAP and HR values at T3, arterial-internal jugular vein bulb oxygen difference [D(a - jv)O 2] at T1, T2, and T3, cerebral oxygen uptake (CEO2) levels, and postawakening agitation scores were lower than those of the control group during the same period (P < 0.05).

IoT-Based Unobtrusive Physical Activity Monitoring System for Predicting Dementia

Mental health-related disorders are common in elderly populations. Among the various mental health disorders, one most significant threat is dementia, and prediction of dementia has become an important issue related to well-being in old age, because the disease progression of dementia can be slowed by early diagnosis and disease control. In this paper, we propose an unobtrusive dementia-prediction system for monitoring physical activities of elderly persons either living alone or as a couple in different house structures, achieved through passive infrared (PIR) motion sensors combined with data processing. The proposed feature extraction algorithm extracts feature values related to physical activities from simple passive infrared sensors located in each room space. We then apply a variety of common popular classification models, including Deep Neural Networks (DNNs), to predict the risk of dementia in a sensor-enabled home. We implemented and validated algorithms on data collected for over a month from 18 participants who were engaged with a variety of living conditions. The proposed system was effective in predicting dementia risk, with up to an 0.99 area under the curve (AUC) using DNN with principal component analysis (PCA) and a quantile transformer scaler. In terms of the result based on leave-one-subject-out (LOSO) analysis, an accuracy of 63.38% was achieved using DNN with PCA and a standard scaler. The proposed methodology is non-invasive and cost-effective, and can be used for a variety of long-term monitoring and early symptom detection systems, helping caregivers provide optimal interventions to elderly individuals at risk for dementia.

Cardiac Arrhythmia Classification Based on One-Dimensional Morphological Features

The electrocardiogram (ECG) is the most commonly used tool for diagnosing cardiovascular diseases. Recently, there have been a number of attempts to classify cardiac arrhythmias using machine learning and deep learning techniques. In this study, we propose a novel method to generate the gray-level co-occurrence matrix (GLCM) and gray-level run-length matrix (GLRLM) from one-dimensional signals. From the GLCM and GLRLM, we extracted morphological features for automatic ECG signal classification. The extracted features were combined with six machine learning algorithms (decision tree, k-nearest neighbor, naïve Bayes, logistic regression, random forest, and XGBoost) to classify cardiac arrhythmias. Experiments were conducted on a 12-lead ECG database collected from Chapman University and Shaoxing People’s Hospital. Of the six machine learning algorithms, combining XGBoost with the proposed features yielded an accuracy of 90.46%, an AUC of 0.982, a sensitivity of 0.892, a precision of 0.900, and an F1 score of 0.895 and presented better results than wavelet features with XGBoost. The experimental results show the effectiveness of the proposed feature extraction algorithm.

A method for eliminating the disturbance of pseudo-textural-direction in ultrasound image feature extraction

Ultrasound imaging is widely used in clinical diagnosis. However, due to the mechanism of B-mode sectorscanning in ultrasound imaging, there exists pseudo-textural-direction (PTD) that is perpendicular to the ultrasonic beam and varies with the change of the beam scanning angle. The existence of PTD reduces the resolution of the image, and brings disturbances to subsequent tasks such as feature extraction, feature screening and segmentation. A method proposed to eliminate the disturbance of PTD consists of three stages. Firstly, a Fourier transform is performed on every superpixel that is a small homogeneous region generated by over-segmentation to obtain a spectrum of the superpixel. Then, ellipse fitting is conducted on the spectrum to detect the deflection angle of textural direction in the superpixel. Finally, the textural direction of the superpixel is corrected by the obtained deflection angle and the co-occurrence matrix texture features are extracted. The effectiveness of the proposed method was verified by segmentation experiments of ultrasound guided images of uterine fibroid acquired from high intensity focused ultrasound (HIFU) therapy based on a split and merge framework. The experimental results were evaluated by quantitative metrics that include true positive (TP), false positive (FP), similarity (SI), Hausdorff distance (HD) and mean absolute distance (MAD). All the metrics were improved in comparison with conventional algorithms when the features screening and segmentation algorithms worked with the proposed feature extraction algorithm. Therefore, the existence of PTD is demonstrated and the proposed method to eliminate its disturbance is feasible.

Demographics and Personality Discovery on Social Media: A Machine Learning Approach

This research proposes a new feature extraction algorithm using aggregated user engagements on social media in order to achieve demographics and personality discovery tasks. Our proposed framework can discover seven essential attributes, including gender identity, age group, residential area, education level, political affiliation, religious belief, and personality type. Multiple feature sets are developed, including comment text, community activity, and hybrid features. Various machine learning algorithms are explored, such as support vector machines, random forest, multi-layer perceptron, and naïve Bayes. An empirical analysis is performed on various aspects, including correctness, robustness, training time, and the class imbalance problem. We obtained the highest prediction performance by using our proposed feature extraction algorithm. The result on personality type prediction was 87.18%. For the demographic attribute prediction task, our feature sets also outperformed the baseline at 98.1% for residential area, 94.7% for education level, 92.1% for gender identity, 91.5% for political affiliation, 60.6% for religious belief, and 52.0% for the age group. Moreover, this paper provides the guideline for the choice of classifiers with appropriate feature sets.

Online detection of class-imbalanced error-related potentials evoked by motor imagery

Objective. Error-related potentials (ErrPs) are spontaneous electroencephalogram signals related to the awareness of erroneous responses within brain domain. ErrPs-based correction mechanisms can be applied to motor imagery-brain–computer interface (MI-BCI) to prevent incorrect actions and ultimately improve the performance of the hybrid BCI. Many studies on ErrPs detection are mostly conducted under offline conditions with poor classification accuracy and the error rates of ErrPs are preset in advance, which is too ideal to apply in realistic applications. In order to solve these problems, a novel method based on adaptive autoregressive (AAR) model and common spatial pattern (CSP) is proposed for ErrPs feature extraction. In addition, an adaptive threshold classification method based spectral regression discriminant analysis (SRDA) is suggested for class-unbalanced ErrPs data to reduce the false positives and false negatives. Approach. As for ErrPs feature extraction, the AAR coefficients in the temporal domain and CSP in the spatial domain are fused. Given that the performance of different subjects’ MI tasks is different but stable, and the samples of ErrPs are class-imbalanced, an adaptive threshold based SRDA is suggested for classification. Two datasets are used in this paper. The open public clinical neuroprosthetics and brain interaction (CNBI) dataset is used to validate the performance of the proposed feature extraction algorithm and the real-time data recorded in our self-designed system is used to validate the performance of the proposed classification algorithm under class-imbalanced situations. Different from the pseudo-random paradigm, the ErrPs signals collected in our experiments are all elicited by four-class of online MI-BCI tasks, and the sample distribution is more natural and suitable for practical tests. Main results. The experimental results on the CNBI dataset show that the average accuracy and false positive rate for ErrPs detection are 94.1% and 8.1%, which outperforms methods using features extracted from a single domain. What’s more, although the ErrPs induction rate is affected by the performance of subjects’ MI-BCI tasks, experimental results on data recorded in the self-designed system prove that the ErrPs classification algorithm based on an adaptive threshold is robust under different ErrPs data distributions. Compared with two other methods, the proposed algorithm has advantages in all three measures which are accuracy, F1-score and false positive rate. Finally, ErrPs detection results were used to prevent wrong actions in a MI-BCI experiment, and it leads to a reduction of the hybrid BCI error rate from 48.9% to 24.3% in online tests. Significance. Both the AAR-CSP fused feature extraction and the adaptive threshold based SRDA classification methods suggested in our work are efficient in improving the ErrPs detection accuracy and reducing the false positives. In addition, by introducing ErrPs to multi-class MI-BCIs, the MI decoding results can be corrected after ErrPs are detected to avoid executing wrong instructions, thereby improving the BCI accuracy and lays the foundation for using MI-BCIs in practical applications.

3D Multi-Camera Calibration with a Fractal Encoded Multi-Shape Target

Robust multi-camera calibration is a fundamental task for all multi-view camera systems, leveraging discreet camera model fitting from sparse target observations. Stereo systems, photogrammetry and light-field arrays have all demonstrated the need for geometrically consistent calibrations to achieve higherlevels of sub-pixel localization accuracy for improved depth estimation. This work presents a calibration target that leverages multi-directional features to achieve improved dense calibrations of camera systems. We begin by presenting a 2D target that uses an encoded feature set, each with 12 bits of uniqueness for flexible patterning and easy identification. These features combine orthogonal sets of straight and circular binary edges, along with Gaussian peaks. Our proposed feature extraction algorithm uses steerable filters for edge localization, and an ellipsoidal peak fitting for the circle center estimation. Feature uniqueness is used for associativity across views, which is combined into a 3D pose graph for nonlinear optimization. Existing camera models are leveraged for intrinsic and extrinsic estimates, demonstrating a reduction in mean re-projection error of for stereo calibration from 0.2 pixels to 0.01 pixels when using a traditional checkerboard and the proposed target respectively.

Anomaly detection for industrial control operations with optimized ABC–SVM and weighted function code correlation analysis

Under the tendency of interconnection and interoperability in Industrial Internet, anomaly detection, which has been widely recognized, has won significant accomplishments in industrial cyber security. However, a crucial issue is how to effectively extract industrial communication features which can accurately and comprehensively describe industrial control operations. Aiming at the function code field in industrial Modbus/TCP communication protocol, this paper proposes a novel feature extraction algorithm based on weighted function code correlation, which not only indicates the contribution of single function code in the whole function code sequence, but also analyzes the correlation of different function codes. In order to design a serviceable detection engine, a dynamic adjusting ABC–SVM (Artificial Bee Colony–Support Vector Machine) anomaly detection model based on double mutations is also developed to identify abnormal behaviors in industrial control communications. The experimental results show that the proposed feature extraction algorithm can effectively reflect the changes of function control behavior in industrial control communications, and the improved ABC–SVM anomaly detection model can strengthen the detection performance by comparing with other anomaly detection engines.

Using deep learning approach and IoT architecture to build the intelligent music recommendation system

First, the local feature extraction of the scale-invariant feature transformation algorithm, the classification excellence of the support vector machine, and the performance of the deep learning-based Fast-RCNN algorithm in the multi-scale feature extraction are analyzed and explained to design an intelligent background music system based on deep learning and Internet of Things (IoT) technology. Then, the intelligent background music system is applied to the Intelligent Home. On this basis, a feature extraction algorithm based on the middle-level feature structure is proposed, which extracts the underlying features of the scene images. Afterward, the critical functional components of the intelligent background music system are explained. Based on the actual operations, an intelligent background music system is designed based on deep learning and IoT. The results show that the recognition rate of indoor scenarios by the middle-level feature construction-based feature extraction algorithm is the highest, which is about 87.6%. The Gabor feature algorithm classifies and identifies the scenarios, and its recognition rate is always around 20%. In the bathroom, the recognition effect of the saliency map feature algorithm is similar to that of the middle-level feature construction-based feature extraction algorithm; however, in the bedroom, the recognition effect of the middle-level feature construction-based feature extraction algorithm is significantly better due to problems such as the lighting and room orientation. The effects of middle-level feature construction-based feature extraction algorithm on the classification and recognition of indoor scenarios are sound. In contrast, the proposed feature extraction algorithm based on deep learning has an optimal effect. The designed and implemented intelligent background music system is stable and effective, which provides a new idea and a new theoretical basis for the future research of intelligent background music system.

Adaptive wavelet thresholding with robust hybrid features for text-independent speaker identification system

The robustness of speaker identification system over additive noise channel is crucial for real-world applications. In speaker identification (SID) systems, the extracted features from each speech frame are an essential factor for building a reliable identification system. For clean environments, the identification system works well; in noisy environments, there is an additive noise, which is affect the system. To eliminate the problem of additive noise and to achieve a high accuracy in speaker identification system a proposed algorithm for feature extraction based on speech enhancement and a combined features is presents. In this paper, a wavelet thresholding pre-processing stage, and feature warping (FW) techniques are used with two combined features named power normalized cepstral coefficients (PNCC) and gammatone frequency cepstral coefficients (GFCC) to improve the identification system robustness against different types of additive noises. Universal Background Model Gaussian Mixture Model (UBM-GMM) is used for features matching between the claim and actual speakers. The results showed performance improvement for the proposed feature extraction algorithm of identification system comparing with conventional features over most types of noises and different SNR ratios.

New Energy Power Prediction Optimization Based on Improved TF-IDF Single Machine Information Feature Extraction

In order to lean management of new energy consumption, based on the characteristics of wide range and complex structure of unstructured data including dispatch and maintenance logs in the power grid system, based on the idea of improving TF-IDF, a DF-IDF data reflecting time slots is proposed Feature extraction algorithm, taking wind power in a power grid in the northern region as an example, extracts keyword and part-of-speech features from unstructured power grid dispatch logs and maintenance data to achieve single-machine state data screening and eliminate irrelevant state data. With the help of the effectiveness of the stand-alone information feature extraction algorithm, the management methods of new energy power prediction optimization are then studied. The results show that in the case of highly complex data, the average accuracy of the DF-IDF-processed feature extraction method is 8% and 7% higher than the traditional TF-IDF method and the One-Hot method. At the same time, it also provides pre-processing methods for stand-alone power prediction calculation, accuracy analysis, and summary station power prediction optimization in new energy power prediction and optimization, improving the rationality, accuracy, and effectiveness of data collaborative processing, and laying new energy lean. Contain the foundation of management level.

Feature Fusion Using Stacked Denoising Auto-Encoder and GBDT for Wi-Fi Fingerprint-Based Indoor Positioning

It may be very difficult to receive the signals from satellite positioning systems due to the existing obstacles in indoor environment. Arising from the popularity of smart phones, Wi-Fi based indoor positioning technology has the advantages with convenient deployment and low hardware cost. In this study, we focused on indoor positioning using Wi-Fi fingerprint data that were collected in shopping malls. Due to the volatility of Wi-Fi signals and the high-dimensional sparseness of fingerprint data, we proposed a feature extraction algorithm, called joint multi-task stacked denoising auto-encoder (JMT-SDAE), aiming at reducing the dimensionality of the original fingerprint data and improving the indoor positioning performance in shopping malls. Furthermore, the features extracted by JMT-SDAE and gradient boosting decision tree (GBDT) were merged to construct a hybrid model, named as JMT-SDAE+GBDT. The experimental results based on 13 location datasets showed that the proposed feature fusion model had better positioning accuracy when compared with other existing positioners, and thus confirmed the effectiveness of our proposed feature extraction algorithm through multi-task learning.

MSdB-NMF: MultiSpectral Document Image Binarization Framework via Non-negative Matrix Factorization Approach.

In this paper, we propose a novel method for Multispectral document image binarization (MSdB) through the Non-negative Matrix Factorization (NMF) approach. We propose a three-step MSdB-NMF framework: i) NMF-based feature extraction algorithm by introducing a new optimization problem; ii) post-processing method iii); apply any existing gray/RGB binarization scheme. In the first step, we extract N features out of B spectral bands (N < B) and their corresponding coefficient matrix. We introduce a novel objective formulation that considers the robustness (related to the noise and various types of degradations) and sparseness (related to the ratio of text pixels versus the background). We employ the multiplicative updating rules to solve the proposed minimization problem and prove the convergence of the proposed feature extraction algorithm. In the next step, we select an appropriate feature vector, equivalently the corresponding coefficient vector. We propose to select it either visually or automatically via a post-processing method, which uses the benchmark binarization methods as baseline. In the last step, we apply some existing binarization methods such as Sauvola and Howe over the selected coefficient vector. Our proposed binarization framework is applicable for any kind of MS or hyperspectral (HS) document image without considering any prior knowledge such as the side information about the spectral bands of MS/HS document image. We evaluate our proposed binarization framework over two MS document image datasets. The experimental results confirm that our proposed framework outperforms several state-of-theart binarization schemes including the winner of the contest in MS-TEx-2015.

GMM-based Automatic Defect Recognition Algorithm for Pressure Vessels Defect Detection through ECPT

In order to realize the automatic identification of pressure vessel defects, an improved adaptive defect recognition feature extraction algorithm through ECPT (Eddy current pulsed thermography) is proposed. The proposed feature extraction algorithm consists of five elements: thermal image data segmentation, variable interval search, probability density function modeling, data classification, and reconstructed image acquisition. The combination of data block selection and variable interval search can reduce the double counting. And the KG-EM (Kmeans-GMM-EM) algorithm is proposed to obtain the Gaussian mixture model corresponding to the classification, and thus the corresponding probability is obtained to classify the TTRs (Transient Thermal Response). The reconstructed thermal image is obtained by the classified TTRs. This method can extract the main information of the image accurately and efficiently. Experimental results are provided to demonstrate their effectiveness.

Multibiometrics System Design Based on Feature Level Fusion

An efficient multimodal biometric system which combines biometric data originated from face, iris and signature biometrics has been presented. Proposed feature extraction algorithm for unimodal and multimodal system has been based on discrete wavelet transform. Among the various biometrics face and iris based human authentication system are proved reliable and efficient. Signature as a behavioral biometrics is very important in financial transaction. Signature has highest variability among all biometrics. This research work proposes an approach to combine signature biometrics with face and iris biometric. Proposed method fuses biometric information originated from face, iris and signature at feature level. Hamming distance based classifier has been used for classifying feature vector as a genuine or imposter. Proposed multibiometrics system has been evaluated on chimeric databases. It has been shown by the reported results that proposed multimodal system outperforms unimodal system performance. Proposed system has been analyzed for recognition rates and error rates. Performance of proposed multimodal system shows improvement in recognition rate and reduction in error.

Integration of morphological preprocessing and fractal based feature extraction with recursive feature elimination for skin lesion types classification

Background and objectiveSkin cancer is the commonest form of cancer in the worldwide population. Non-invasive and non-contact imaging modalities are being used for the screening of melanoma and other cutaneous malignancies to endorse early detection and prevention of the disease. Traditionally it has been a problem for medical personnel to differentiate melanoma, dysplastic nevi and basal cell carcinoma (BCC) diseases from one another due to the confusing appearance and similarity in the characteristics of the pigmented lesions. The paper reports an integrated method developed for identifying these skin diseases from the dermoscopic images. MethodsThe proposed integrated computer-aided method has been employed for the identification of each of these diseases using recursive feature elimination (RFE) based layered structured multiclass image classification technique. Prior to the classification, different quantitative features have been extracted by analyzing the shape, the border irregularity, the texture and the color of the skin lesions, using different image processing tools. Primarily, a combination of gray level co-occurrence matrix (GLCM) and a proposed fractal-based regional texture analysis (FRTA) algorithm has been used for the quantification of textural information. The performance of the framework has been evaluated using a layered structure classification model using support vector machine (SVM) classifier with radial basis function (RBF). ResultsThe performance of the morphological skin lesion segmentation algorithm has been evaluated by estimating the pixel level sensitivity (Sen) of 0.9172, 0.9788 specificity (Spec), 0.9521 accuracy (ACU), along with the image similarity measuring indices as Jaccard similarity index (JSI) of 0.8562 and Dice similarity coefficient (DSC) of 0.9142 with respect to the corresponding ground truth (GT) images. The quantitative features extracted from the proposed feature extraction algorithms have been employed for the proposed multi-class skin disease identification. The proposed layered structure identifies all the three classes of skin diseases with a highly acceptable classification accuracy of 98.99%, 97.54% and 99.65% for melanoma, dysplastic nevi and BCC respectively. ConclusionTo overcome the difficulties of proper diagnosis of diseases based on visual evaluation, the proposed integrated system plays an important role by quantifying the effective features and identifying the diseases with higher degree of accuracy. This combined approach of quantitative and qualitative analysis not only increases the diagnostic accuracy, but also provides some important information not obtainable from qualitative assessment alone.

A New Feature Extraction Algorithm to Extract Differentiate Information and Improve KNN-based Model Accuracy on Aquaculture Dataset

In the world of aquaculture, understanding the condition of a pond is very important for a farmer in deciding which action should they take to prevent any bad condition occurred. Condition of a pond can be justified by measuring plenty of water parameters which can be divided into 3 categories that are physical, chemical and biological. The physical parameter is any physical quantity that can be measured in the pond. The chemical parameter is any kind of chemical substances that are dissolved in water. The biological parameter is any organic matter that lives in water. However, all of these parameters are not so distinguishable in representing the condition of a pond. Therefore, the farmer experience difficulties in justifying the condition and taking proper action to their pond. Even with the help of the K-Nearest Neighbors (KNN) algorithm combined with grid search optimization to model the data, the result is still not satisfying where the model only achieve accuracy of 0.701 in leave one out validation. To overcome this problem, a kind of feature extraction algorithm is needed to extract more information and make the data become more differentiate in representing the condition of the pond. With the help of our proposed feature extraction algorithm, optimized KNN can model the data easier and achieve higher accuracy. From the experiment results, the proposed feature extraction algorithm gives an impressive performance where it increases the accuracy to 0.741. A comparison with other feature extraction algorithms such as Principal Component Analysis (PCA), Non-negative Matrix Factorization (NMF), and Singular Value Decomposition (SVD) is also conducted to validate how good the proposed feature extraction algorithm is. As a result, the proposed algorithm is surpassing the other algorithms which only achieve the accuracy of 0.707, 0.718, and 0.718, respectively.

A location-aware feature extraction algorithm for image recognition in mobile edge computing.

With the explosive growth of mobile devices, it is feasible to deploy image recognition applications on mobile devices to provide image recognition services. However, traditional mobile cloud computing architecture cannot meet the demands of real time response and high accuracy since users require to upload raw images to the remote central cloud servers. The emerging architecture, Mobile Edge Computing (MEC) deploys small scale servers at the edge of the network, which can provide computing and storage resources for image recognition applications. To this end, in this paper, we aim to use the MEC architecture to provide image recognition service. Moreover, in order to guarantee the real time response and high accuracy, we also provide a feature extraction algorithm to extract discriminative features from the raw image to improve the accuracy of the image recognition applications. In doing so, the response time can be further reduced and the accuracy can be improved. The experimental results show that the combination between MEC architecture and the proposed feature extraction algorithm not only can greatly reduce the response time, but also improve the accuracy of the image recognition applications.