$20 Bonus + 25% OFF CLAIM OFFER
Place Your Order With Us Today And Go Stress-Free
The current study examines the effects of preprocessing methods, are Daubechies wavelets (db1, db2, db3) and the Fast Fourier Transform , on the performance of image classification tasks. The main goals include evaluating the efficacy of these methods for preprocessing and conducting a comparative analysis of their performance in face identification, gender classification, and unsupervised clustering tasks. Preprocessing methods are of significant importance in the effort of improving image categorization accuracy.
The issue statement is originally provided, emphasising the crucial significance of preprocessing in the setting of image classification. Additionally, the scope of the research is explicitly defined. In the process of data preparation, an extensive collection comprising a wide range of facial photos is utilized. For the consistency, this dataset goes through an extensive cleaning, normalization, and scaling procedure.
Daubechie wavelets and the Fast Fourier Transform are used in the feature extraction process to transform photos into useful feature representations.
Many classification approaches for including Support Vector Machines (SVM), Convolutional Neural Networks (CNN), and k-nearest Neighbors, are used to recognize faces and classify genders. Specifically, support vector machines are studied using different kernel functions, convolutional neural networks are customized with appropriate network structures, and k-nearest neighbors are improved for optimal nearest neighbor count and distance measurement techniques.
Unsupervised clustering methods, such as K-Means, Hierarchical Clustering, DBSCAN, and Gaussian Mixture Models (GMM), are utilized to detect inherent patterns in facial images. The determination of the most suitable number of clusters is affected by various criteria, including the Elbow Method, Silhouette Score, dendrogram visualization, and information criteria such as BIC and AIC.
The main results of this study provide valuable insights into the advantages and drawbacks of using Daubechies wavelets and Fast Fourier Transform as preprocessing methodologies in the context of image categorization. The accurate assessment of their influence can be achieved by performing comparative evaluations across many measures, including as accuracy, precision, recall, and Fβ-score.
Preprocessing approaches serve a crucial role in helping the conversion of raw image data into more appropriate representations, consequently enhancing the accuracy and efficiency of classification algorithms employed in image classification tasks. The objective of this work is to evaluate the efficacy of Daubechie Wavelets are (DB1, DB2, and DB3) and Fast Fourier Transform in enhancing the efficiency of photo classification tasks, using two distinct preprocessing methods.
Spanning a wide variety of applications including face identification, gender categorization and others, considering image classification is essential for not only computer vision but also pattern recognition and image processing.
The quality of classifying an image is heavily dependent on the precision with which its features are extracted.
Preprocessing is very important in image categorization. Interpretation of meaningful patterns presents classifiers with great obstacles, since raw visual data is highly complex and high dimensional. [6].
There are preprocessing methodologies which not only extract distinctive features but also reduce the dimensionality of the data, making it easier to classify. The Daubechies wavelets are particularly famous for their excellent information collection and representation abilities in both time and frequency domains. Most people accept the effectiveness of Fast Fourier Transform in analyzing frequency data.
Despite the differences between these two preprocessing methods, their effectiveness is similar [2]. The study aims to measure the effectiveness of a number of preprocessing methods in order to get better insight into whether they are suitable for different image classification problems.
This paper's aim is to determine how different preprocessing techniques affect the accuracy of categorizing an image. With this objective in mind, the three different activities to be reviewed will have their own specific needs and problems. The purpose of this study is to evaluate and compare the efficacy of preprocessing technique.
Also Read - Programming Assignment Help
Because methods for preprocessing images are an important link in the process of image classification, their role has been recognized by many researchers. A great many attempts have been made to realize this improvement [3].
Its ability to extract relevant information quickly and efficiently from images has attracted much interest in the academic community for wavelet transforms, as well as Fourier transforms. This section provides a thorough review of past research on preprocessing methodologies. In particular, it evaluates the pros and cons associated with each approach.
Since wavelet transforms are able to simultaneously evaluate information in immediate and frequency domains moreover, there has been a great deal of study on them in the field of image pre-processing[2,3]. The extent to which they can perform well at denoising images or feature extraction such as db1 anddb 2 were particularly welcome for their symmetrical characteristics. These transformations have been used in a variety of image classification applications such as texture identification and face recognition.
In the framework of multiresolution analysis, wavelets enable us to capture different characteristics at different scales within images.
Daubechies wavelets can provide a focused description of features, and for this reason the important details in an image are preserved [5].
Power Compaction: In wavelet transforms, a large portion of the energy in an image ends up highly concentrated at very few coefficients. Wavelet transformations have the property enabling efficient dimensionality reduction.
The Fourier transform is a frequently used preprocessing method in the domain of image analysis. The method involves transforming an image from the spatial domain to the frequency domain, revealing information about the dominant frequencies and orientations present in the image. Preliminary inquiries employing Fourier analysis as a method of categorization. This method has been subsequently used for subsequent investigations within the domains of face recognition and identification.
Fourier transformations are successful in extracting frequency-related information, making them well-suited for applications that need a focus on texture or periodicity .
The computational efficiency of calculating Fourier coefficients can be significantly improved, particularly through the utilization of the Fast Fourier Transform (FFT) technique.
The utilization of Fourier transforms offers an extensive representation of the image, hence presenting potential benefits in specific situations.
In this research, a comprehensive dataset containing facial photos sourced from various origins was used to perform face identification, gender classification, and unsupervised clustering tasks. The dataset consists of photos showing people from various age groups, ethnic backgrounds, and genders. It includes the subsequent essential attributes, which are:
The collection provided comprises high-resolution facial images of humans, with each image accompanied by a unique identity that corresponds to a single person. The dataset functions as the main source of data for the face identification challenge. The collection is of significant size, contains a multitude of photos, and includes an array of positions, expressions, and lighting situations.
A dataset that concentrated on gender categorization was developed from the larger face identification dataset. This particular selection includes an equitable distribution of both male and female participants [8, 9]. The gender of each participant is indicated with a label accompanying the respective image. With the provided dataset, one can evaluate how well each of these methods preprocesses data in order to distinguish male and female facial images.
In this study the unsupervised clustering dataset with no associated labels consisted entirely of facial images. The principal use of this data set is to allow analysis, through an unsupervised method,of the basic structure and groupings in facial characteristics.
They are important in terms of the accuracy and suitability of the data for further analysis. The subsequent procedures were executed:
The facial image dataset underwent massive data cleaning work to cut out all photos either corrupted or incomplete. Moreover, all unnecessary metadata and annotations were deleted in order to keep the quality of data intact.
To deal with changes in lighting conditions and pixel intensity, normalization was performed on the photos. The process included adjusting the brightness and contrast levels of images to maintain uniformity.
To resize means to adjust the dimensions of all photos to a common resolution. This was to increase computational efficiency during the preprocessing and classification stages.
Utilizing a face detection algorithm, the facial regions in each image were localized and extracted [8, 9]. It used this cutoff point to ensure that all face aspects relevant only for analysis were excluded as outputs, making the initial processing stage easier.
Using the extraction process, facial images have been converted into a form that can be used in classification and clustering. Wavelet preprocessing techniques (db1, db2 and so on) were used to compute wavelet coefficients for each image.
The purpose of this discourse is to explain the theoretical foundations behind Daubechies wavelets (db1, db2 and db3) as well as for Fast Fourier Transform.
A set of wavelets which is often used in the areas of signal processing and image compression are known as daubechies. They include db1, db 2 and db3 [9].
These properties of compact support and orthogonality, led the Daubechies wavelets to capture attention in the area of wavelet transformations. The activities of image and signal processing are generally widely used. The underlying principle of wavelet transforms is to express a signal or image in terms of both its temporal structure and frequency nature. The Daubechies wavelets, which are usually written as db1, db2 and so on according to their prescribed filter coefficients.
The Haar wavelet are represented as db1, is a fundamental wavelet transform consisted of two coefficients. The signal is divided into two segments by calculating the difference in variance and mean values between successive readings. The methodology shows efficacy in discerning distinct limits within a image.
Four different components can be used to identify a waveform. This wave is primarily intended for accurately capturing a signal's lower-frequency and high-frequency components, resulting in a representation of the signal. The DB3 wavelet can be improved by adding a combination of six coefficients, which will lead to better frequency localization. This type of approach is frequently utilized for the objective of analyzing complex data.
This study examines at the prominent frequencies and their corresponding orientations of an image. A commonly used method in image processing for swiftly converting a signal's or image's spatial representation into the frequency domain is the Fast Fourier Transform[8]. The basis of it is the idea that each given signal can be divided up into separate sinusoid components, each of which has a different frequency and amplitude.
A Various commonly used classification approaches were used to perform the tasks of facial identification and gender classification. This study offers a comprehensive examination of the different categorization methods used in the field, along with an assessment of their distinct attributes and arrangements.
The study is utilized various kernels for including linear, polynomial, and radial basis functions to examine different decision boundaries.
In the tests are manipulated the regularization value, abbreviated as C, in order to achieve a balance between optimizing the margin and minimizing classification errors.
These parameters were adjusted by using grid search and cross-validation to determine the ideal combination for each kernel.
The study of architecture includes the design, planning, and construction of buildings and other structures. The input layer includes convolutional and pooling layers, which are then followed by fully connected layers. The activation functions frequently employed in neural networks include Rectified Linear Unit (ReLU) for the layers that are hidden and Softmax for the output layer.
The learning rate is modified in order to control the rate at which convergence occurs.
The batch size is calculated by considering the amount of accessible memory.
The selection of the number of periods has been driven by the objective of minimizing the issues of underfitting and overfitting. The optimization algorithms used in the study included several variations of stochastic gradient descent, Adam, and RMSprop.
It conducted experiments by using various values of k with the goal of determining the ideal number of nearest neighbors. The selection of a distance metric for measuring similarity between data points is dependent upon the specific dataset, with the Euclidean distance being one possible option among others. The weighting scheme for consideration includes both uniform and distance-weighted k-NN approaches.
Regularization techniques like L1 and L2 regularization are frequently used in machine learning and statistical modeling to decrease the risk of excessive overfitting. To efficiently regulate and control the degree of influence of regularization, the regularization strength is modified. In the experiment, the maximum dimension of the model was limited, and the number of decision trees in the random forest was changed. [9,10].
Decision tree algorithms can be selected using entropy and the Gini impurity. In order to optimize the ratio of variance to bias, a variety of parameters and configurations are used in gradient boosting. The maximum depth is controlled to reduce the chance of overfitting, and the learning rate is altered to control the impact of each estimator.
The Various recognized methods of clustering are used to identify inherent patterns in facial features while guaranteeing the presence of particular instructions [9,10]. It provides a thorough description of the algorithms used for unsupervised clustering and explains how various considerations are taken into account when calculating the most suitable number of clusters. Grouping data points. to a group of computing techniques, unsupervised clustering algorithms are.
K-Means is a center based clustering technique that attempts to divide data into k clusters. The mean, or centroid, of the data points corresponding to each cluster defines that particular one. The factors impacting the determination of the optimal number of clusters are as follows: This study employed multiple methods to determine the number of clusters that are most apt, such as using the elbow method and silhouette score. The Elbow Method is an approach that attempts to find the inflection point at which the rate of change in WCSS begins to slow. This score, called a silhouette measure quantifies the level of cohesion and separation within clusters.
A hierarchical clustering algorithm makes a hierarchial structure of clusters by continuing the process downward through merging or splitting up groups, according to a linkage criteria such as single-linkage complete-or average link.
The hierarchical dendrogram was presented and the determination of number of clusters done by recognizing the natural breaks indicated in how they branch on the tree [10]
This study used both wavelet transformations and the Fast Fourier Transform (FFT) method to preprocess a dataset of facial images. A cell array 'images' is created to contain a set of facial images. Those graphic illustrations have a systematic format. A grid structure is employed. In this grid, the individual cells are used to accommodate one image and its accompanying title.
The method of depicting and perceiving images with visual mediums. So the code then depicts in a visual fashion the array of images imported into it by way of complications, thus displaying them in a sortof grid form. This is a particularly important stage in understanding the dataset and its parts.
Uncorking imagesSetting the stageThe code following specifies a path setting referring to 'pgm_image_dir', where uncropped . pgm photos are placed. This program downloads a batch of image files from the chosen directory and proceeds to create an image matrix called 'image_matrix'. The purpose of this matrix is to store flattened versions of the photos. Each image is read, flattened and becomes a column in the matrix. This procedure is often used as a preliminary to data analysis.
The algorithm produces labels for each image, which may represent separate identities: especially in the case of facial images [4]. The program preserves both the matrix of flattened image and its corresponding labels, in .mat file format. This is an important aspect in subsequent tasks such as classification and grouping.
Data Splitting: The dataset is divided into two separate subsets, the training set and the testing set. The training set is 70 % of the entire dataset, and testing makes up the remaining 30 %. The process of randomization guarantees that the allocation of subjects or treatments is both representative and capable of being reproduced.
The utilization of wavelets and fast Fourier transform (FFT) at the preprocessing stage. The primary emphasis of this code segment pertains to the preparation of data.
The application of wavelet transform and Fast Fourier Transform (FFT) is employed on both the training and testing data. The decomposition of each image is performed using the 2D wavelet transform, specifically the 'wavedec2' function. Subsequently, the Fast Fourier Transform (FFT) is computed on the resulting wavelet coefficients.
Figure 1 : Array of images
Figure 2 : Singular values
The Yale Faces dataset was used to conduct a face classification employment a k-nearest neighbors (KNN) classifier.[6]
The code initiates the loading process of the labels for the Yale Faces dataset from a specific file named 'yalefaces_labels.mat'. The aforementioned labels are assumed to symbolize the identities or classifications linked to each individual image.
The process of selecting labels involves choosing only the initial 30 labels, which match to the first 30 images inside the collection. This implies that the primary emphasis of the classification task is in the examination of these initial images.
The dataset has been divided into training and testing sets with a predetermined ratio (80% for training and 20% for testing). The determination of the total number of samples and the number of training samples is conducted. Next, the indices of the samples are randomized to ensure a random distribution.
Data Preparation: The provided code is responsible for the preparation of both the training and testing data sets. A predetermined quantity of principle component is chosen from the data that has been converted using the singular value decomposition (SVD) technique. The primary components function as characteristics for the classification process. The training and testing datasets are generated by utilizing the primary components that have been identified.
The process of training a K-Nearest Neighbors (KNN) classifier involves several steps. First, the dataset needs to be prepared by splitting it into a training set and a test set. The training A K-nearest neighbors (KNN) classifier is instantiated with a predetermined number of nearest neighbors, in this case, k=5.
Figure 3: PCA
Figure 4 : Singular value spectrum
Figure 5 :Rotated Images
Figure 6 :Test face classification
A gender categorization job is conducted on the Yale Faces dataset with a k-nearest neighbors (KNN) classifier.
The code initiates the process by importing the labels associated with the Yale Faces dataset from a designated file named 'yalefaces_labels.mat'. It is presumed that these labels correspond to the gender of each participant inside the dataset.[8]
The Conceptualization of Gender Categories: Gender labels are assigned to each image inside the dataset. In this particular instance, the numerical value '1' is employed to symbolize the male gender, whereas the numerical value '2' is used to symbolize the female gender.
Data Partitioning: The dataset is segregated into two distinct sets, namely the training set and the testing set, with a predetermined partition ratio of 80% for training and 20% for testing. The determination of the total number of samples and the number of training samples is conducted. In order to assure unpredictability in the data split, the indices of the samples are randomly swapped.
Data Preparation: The code does data preparation by picking a predetermined number of primary components (specifically, 12 components) from the data that has been processed using Singular Value Decomposition in a prior step. The major components in question function as attributes for the purpose of gender classification.
The process of training a KNN classifier involves instantiating the classifier with a predetermined number of nearest neighbors, in this case, k=5. The training dataset and its labels will be used to train the K-Nearest Neighbors (KNN) model for the purpose of gender classification.
Figure 7 : Gender classification
The preprocessed image data is subjected to unsupervised clustering using the K-Means method.
The process of loading data and labels is initiated. The code begins with the line import 'yalefaces_images.mat' ./data which brings us into possession of both the part-blinded image data imported from the file yalefaces_images. mat and try labels accompanying these images, all due to presence in this same parent folder, are contained in another two files. Those labels in question are often tools used to express the individual personalities of people within that collection.
The determination of the best number of clusters: [5].
In the K-Means algorithm, 15 is assigned as the value for number of clusters. It is hypothesized that this numerical value represents the number of different species within the collection. The K-Means algorithm will try to split the data set into 15 separate groups, each group representing a different person.
The data can be normalised through a procedure.
In this instance, the code provides one method of normalizing data by applying z-score standardization. By means of normalization techniques, the K-Means algorithm can be made even more effective, with all features reduced to a uniform scale. In this particular case, the data is normalized using principle components obtained from singular value decomposition (SVD).
The K-Means clustering process is carried out on the normalized dataset. A fixed number of clusters (15) is used in the 'kmeans' function, and the property Replicates is set to 10, causing K-Means with multiple first starts. This reduces drastically the likelihood that one falls into a local minimum point.
That is why we display the results of the K-Means clustering algorithm on a three-dimensional scatter plot. Each data point is plotted in three-dimensional space along the first three principal components, representing an individual's face.
Cluster analysis is used to further understand the results of clustering. To do this, the algorithm goes through each cluster and finds which individuals belong to it. The output includes the numerical identifier of each cluster as well as the indices of persons assigned to each respective cluster.
Figure 8 : Unsupervised learning
Figure 9 : Classification after preprocessing
Figure 10 : Test 3 outcome
The results of three different tests are compared are as follows:
The results of this test 1 were stored in the variables test1_accuracy and test1_confusion_matrix.
The data from the face classification test is stored in variables called test1_accuracy and test1_confusionmatrix. This study has involved classifying facial images into discrete groups according to the identity of each[4,5].
The accuracy figure indicates what proportion of faces were correctly classified, but the confusion matrix presents a detailed dissection of true positives, true negatives, false positives and false negatives.
Test2 focused on gender classification. The results of this test were entered into the variables test2_accuracy and test2_confusion matrix.
Test 2 The accuracy and confusion matrix generated by the gender categorization test are stored in variables of similar name to those used for Test 1,test2_accuracy and
test2_confusionmatrix. That is, this study involved classifying facial images into separate male and female groups.
The accuracy metric represents the proportion of gender labels which were correctly assigned, and offers a comprehensive viewpoint.
Test 3 : Unsupervised Learning using K-Means Clustering
The experiment involved the use of unsupervised learning by K-Means clustering to the preprocessed face data in order to categorize faces that share similarities.
The Within-Cluster Sum of Squares (WCSS) for each cluster, hence offering valuable insights into the level of compactness exhibited by the clusters.
The outcomes of Test 3 do not pertain to accuracy or confusion matrices, but instead involve WCSS values, which serve as a means to assess the clustering's quality [7].
Test Results: Subsequently, the code exhibits the outcomes of the three tests, furnishing a concise overview of the performance of each individual test.
The accuracy values and confusion matrices for Test 1 and Test 2 are crucial components in the context of supervised classification tasks.
Test 3 presents the Within-Cluster Sum of Squares values for each cluster, a crucial statistic used to assess the efficacy of unsupervised clustering.
Figure 11 : Comparison result
Figure 12 : Test 1 and 2 comparison result
Figure 13 : Test 3 comparison result
The code performs several tasks related to data augmentation, dimensionality reduction, and binary classification using Convolutional Neural Networks (CNN) and Support Vector Machines (SVM).
The code starts with data augmentation, specifically rotating images [3,6]. The imrotate function rotates 30 degrees the twelve female images. The images are then rotated and flattened, followed by Singular Value Decomposition (SVD).
The code uses the svd function to compute the SVD of these augmented, rotated images. We use the singular values to plot rank of SVD for both rotated and augmented images. This step is important for understanding the form of the data, and also reduces dimensions.
A desired number of principal components (num_principal_components) is selected for data reduction. The chosen principal components are then used to reconstruct the recovered images [2]. The images are displayed, and Discrete Wavelet Transform (DWT) is applied to the image.
Figure 14: Singular Values of Rotated and Augmented Images
For the binary classification of two classes, a CNN model is defined. It contains convolutional layers with max-pooling, fully connected layers and a softmax output layer. It is assumed that the training labels cover all samples and data are prepared accordingly. The CNN model is trained with stochastic gradient descent with momentum ('sgdm') and other training options.
Figure 15: Performance and Epochs of CNN & SVM
Evaluation of CNN:
Predictions (cnn_predictions) are made with the trained CNN model. The accuracy and confusion matrix of the CNN model are calculated and displayed. Printing out the model's performance metrics, including accuracy and the confusion matrix [10].
Support Vector Machine (SVM) for Binary Classification:
A binary classification model based on Support Vector Machines (SVM) and Error-Correcting Output Codes (ECOC), is trained. The predictions are made using the trained SVM model and calculated performance metrics (accuracy, confusion matrix) are displayed.
Statistical Analysis:
Performance tests: The performance of the CNN and SVM models are compared by using a statistical analysis (a two-sample t test--ttest2). The t-statistic, p value and confidence intervals are shown. It is from these values that insights might be drawn on whether there was a significant difference of performance between the two models.
Visualization:
Bar plots are used to represent the results of both CNN and SVM models [9]. True labels and predicted labels are plotted side by side for comparison.
The code, in a nutshell: data augmentation; dimensionality reduction via SVD decomposition of the feature matrix; training and assessment of binary classifier generated by CNN model (classification involves but is not limited to sklearn); Training and evaluation for an SVM based binary classifier. Comparative statistical analysis between two models The visualizations help us better understand and compare the results of these two models. Each step of the process is well commentated, and makes for easily understanding code.
The results of three separate examinations are as follows: facial categorization (Test 1), gender categorization (Test 2), and unsupervised learning employing K-Means clustering (Test 3).
The first test is called supervised face classification, which means as far as the code itself goes simply that it computes and stores an accuracy value (the percentage of correct choices) along with a confusion matrix to distinguish between classes by their identity. The accuracy measure expresses the percentage of faces that have been correctly identified. The assessment has great value in evaluating the model's ability to separate out various identities within this dataset [4].
For the second test, particularly designed to evaluate gender classification, an accuracy and confusion matrix is calculated and retained by the algorithm. These criteria allow faces to be divided into two different sex categories-male or female. The measure of rightness concerns the ratio among gender labels correctly sorted. For assessing the model's ability to distinguish between facial features identified with males and females, this assessment is helpful.
Lastly, the third test makes use of unsupervised learning through K-Means clustering. The evaluation of the quality of clustering is achieved by computing, for each cluster, a Within-Cluster Sum of Squares (WCSS). The inside-cluster sum of squares (W CSS) measure gives us some idea about how tight data points are clustered within each cluster, the lower values indicating tighter clustering.
When analyzing the findings, it is imperative to take into account the subsequent observations:
The correctness of the model is proven in its ability to accurately recognize unique individuals among a group of facial images. A matrix is used to measure how well the different people can be distinguished. Accuracy and a well-balanced confusion matrix identify effective face recognition [2, 8]
The second examination focuses on sex classification; it seeks to provide a measure of how well the model can distinguish between male and female faces. Important measures by which gender categorization performance is evaluated are accuracy and confusion matrix.
The third test involves the use of unsupervised K-Means clustering. WCSS values are used to measure the efficiency of an algorithm for clustering together similar facial features.
To summarize, the code provided considerable analysis of face data: this included classifying faces as male or female and carrying out unsupervised clustering. Every test has its own specialties, not only to evaluate the model's performance but also from which we can draw in-depth insights about our data.
In this first face classification test (Test 1), some important conclusions on the ability of the model to correctly distinguish different identities were obtained. Information about the ability of the model to differentiate persons was also provided by these tests, which is particularly important when applications like facial recognition systems are involved. The confusion matrix also provided a way to assess the model's performance in this area.
The second test (Test 2) was namely gender classification. It gauged whether or not a model could classify faces into different genders categories [6]. For example, we measure a model for recognizing male and female faces by accuracy metrics and confusion matrices. Therefore, this analysis offered many pointers on how to proceed with gender-based analyses [8].
The unsupervised clustering test (Test 3 in this example) uses a K-Means algorithm to find the hidden from among face datasets. To evaluate the quality of clustering, we used Within-Cluster Sum of Squares. Values lower within-cluster sum of squares show that this model can yout groups together faces similar to one another without using labeled data.
Optus data breach case study | Introduction to webtechnology using Javascript | Medibank Data Breach Case Study |
[1] Gómez-Echavarría, A., Ugarte, J.P. and Tobón, C., 2020. The fractional Fourier transform as a biomedical signal and image processing tool: A review. Biocybernetics and Biomedical Engineering, 40(3), pp.1081-1093.
[2] Gokulalakshmi, A., Karthik, S., Karthikeyan, N. and Kavitha, M.S., 2020. ICM-BTD: improved classification model for brain tumor diagnosis using discrete wavelet transform-based feature extraction and SVM classifier. Soft Computing, 24, pp.18599-18609.
[3] Kumar, D.M., Satyanarayana, D. and Prasad, M.G., 2021. An improved Gabor wavelet transform and rough K-means clustering algorithm for MRI brain tumor image segmentation. Multimedia Tools and Applications, 80, pp.6939-6957.
[4] Agustika, D.K., Mercuriani, I., Purnomo, C.W., Hartono, S., Triyana, K., Iliescu, D.D. and Leeson, M.S., 2022. Fourier transform infrared spectrum pre-processing technique selection for detecting PYLCV-infected chilli plants. Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, 278, p.121339.
[5] Al-Azzawi, A., 2021. Fourier transform based epileptic seizure features classification using scalp electrical measurements using KNN and SVM (Master's thesis, Altınbaş Üniversitesi).
[6] Mahmood, B.A. and Kurnaz, S., 2023. An investigational FW-MPM-LSTM approach for face recognition using defective data. Image and Vision Computing, 132, p.104644.
[7] Strömbergsson, D., Marklund, P., Berglund, K. and Larsson, P.E., 2020. Bearing monitoring in the wind turbine drivetrain: A comparative study of the FFT and wavelet transforms. Wind Energy, 23(6), pp.1381-1393.
[8] Muqeet, M.A. and Holambe, R.S., 2019. Local binary patterns based on directional wavelet transform for expression and pose-invariant face recognition. Applied Computing and Informatics, 15(2), pp.163-171.
[9] Poornima, S., Sripriya, N., Vijayalakshmi, B. and Vishnupriya, P., 2017, January. Attendance monitoring system using facial recognition with audio output and gender classification. In 2017 International Conference on Computer, Communication and Signal Processing (ICCCSP) (pp. 1-5). IEEE.
[10] Thangaraj, R., Pandiyan, P., Pavithra, T., Manavalasundaram, V.K., Sivaramakrishnan, R. and Kaliappan, V.K., 2021, October. Deep Learning based Real-Time Face Detection and Gender Classification using OpenCV and Inception v3. In 2021 International Conference on Advancements in Electrical, Electronics, Communication, Computing and Automation (ICAECA) (pp. 1-5). IEEE.
Are you confident that you will achieve the grade? Our best Expert will help you improve your grade
Order Now