Academia.eduAcademia.edu

Figure 13 – uploaded by Surasak Mungsing

See full PDFdownloadDownload figure
Fig. 13 Shows an Average of Error Mean and Standard Deviation Fig. 12 Compares a Difference in Error of the Actual Answer with ANFIS Calculation. It Shows a Little Error.

Figure 13 Shows an Average of Error Mean and Standard Deviation Fig. 12 Compares a Difference in Error of the Actual Answer with ANFIS Calculation. It Shows a Little Error.

Related Figures (19)

Fig. 1 Shows a Main Component of Face Recognition Third, Face recognition inputting a specific data of specific face features from the second Feature Extraction to compare with a specific data of face features then make a result assessment on the closest face features as shown in Fig. 1.
Fig. 1 Shows a Main Component of Face Recognition Third, Face recognition inputting a specific data of specific face features from the second Feature Extraction to compare with a specific data of face features then make a result assessment on the closest face features as shown in Fig. 1.
Fig. 2 Fuzzy Structure Model Sugeno Fuzzy 1. Sugeno Fuzzy structure or TS: Takagi- Sugeno Fuzzy is shown in Fig. 2. The data sample input in 2 dimensions is brought to be assessed in function and Input Membership Function. The result to be used to determine Rule Weight or Firing Strength for the variable assessment from an output and a relationship of an equation (1) can be exhibited.
Fig. 2 Fuzzy Structure Model Sugeno Fuzzy 1. Sugeno Fuzzy structure or TS: Takagi- Sugeno Fuzzy is shown in Fig. 2. The data sample input in 2 dimensions is brought to be assessed in function and Input Membership Function. The result to be used to determine Rule Weight or Firing Strength for the variable assessment from an output and a relationship of an equation (1) can be exhibited.
The first layer or Input layer determined Input data to be represented by x with the
The first layer or Input layer determined Input data to be represented by x with the
Fig. 4 A Structure of Brain Neural Network Lippmann (1987), Chester (1993) and Kireetoh (1995) explained that Neural was similar to human nerves accepting an incoming input and was stimulated. Each cell consisted of the end of nerve stream called “Dendrite”, the incoming data input. While sending data to nerve stream is called “Axon”, an output data from cells. These cells worked with Electrochemical system when motivated by a stimulant or by other cells. A neura stream ran through Dendrite to Nucleus as it decided whether to motivate other cells or not. If the stream was stronger, nucleus would motivate other cells further through its Axon. The final result of the same motivation or similar attributes would give an absolute value. The Artificial Neural Network is an Artificial Intelligence created to imitate human’s Neural system (Lippmann, 1987; Caudill and Butler, 1990; Klimasauskas, 1993; Medsker et al., 1993). The Artificial Neural Network system works simailarly to human’s Neural system, learning from past experience and being able to find an answer even though data input is an error or incomplete, by finding a method to solve a problem from its past experience and being able to further develop to correspond with human.
Fig. 4 A Structure of Brain Neural Network Lippmann (1987), Chester (1993) and Kireetoh (1995) explained that Neural was similar to human nerves accepting an incoming input and was stimulated. Each cell consisted of the end of nerve stream called “Dendrite”, the incoming data input. While sending data to nerve stream is called “Axon”, an output data from cells. These cells worked with Electrochemical system when motivated by a stimulant or by other cells. A neura stream ran through Dendrite to Nucleus as it decided whether to motivate other cells or not. If the stream was stronger, nucleus would motivate other cells further through its Axon. The final result of the same motivation or similar attributes would give an absolute value. The Artificial Neural Network is an Artificial Intelligence created to imitate human’s Neural system (Lippmann, 1987; Caudill and Butler, 1990; Klimasauskas, 1993; Medsker et al., 1993). The Artificial Neural Network system works simailarly to human’s Neural system, learning from past experience and being able to find an answer even though data input is an error or incomplete, by finding a method to solve a problem from its past experience and being able to further develop to correspond with human.
Fig. 5 A Working Process in Artificial Neural Network in Sub Nodes Summation function is a function to find an average weight of every connecting Node. The procedures are to input data in each Node and to multiply by a weight of every Node and sum all Nodes together as shown in Fig. 5. Whereas Transfer function is a relationship between a level of stimulation within a Node (N) under conditions: 1) continuous and 2) value of function Sigmoy will increase when N increases (Smith, 1993).
Fig. 5 A Working Process in Artificial Neural Network in Sub Nodes Summation function is a function to find an average weight of every connecting Node. The procedures are to input data in each Node and to multiply by a weight of every Node and sum all Nodes together as shown in Fig. 5. Whereas Transfer function is a relationship between a level of stimulation within a Node (N) under conditions: 1) continuous and 2) value of function Sigmoy will increase when N increases (Smith, 1993).
Robust Face Recognition Technique with ANFIS in Compared with Feedforward Backpropagation Neural Network Using Specific Features of Wavelet Transform and HOG
Robust Face Recognition Technique with ANFIS in Compared with Feedforward Backpropagation Neural Network Using Specific Features of Wavelet Transform and HOG
Stop training Network can be performed in two ways: 1) determine training cycle (Epochs) and 2) determine an acceptable value of error. Bhokha, (1998) Carpenter (1993) suggested to determine a number of a training cycle at 20,000 - 100,000. Another method is to determine a deviation between actual data and a result that the Network can calculate (Khan et al., 1993). A caution is that training a Network too long will make a problem called over fitting as shown in Fig. 8, in which the Network can learn to achieve the least error result. When testing with a testing set it appears that it cannot give a good the result (Bhokha, 1998). Samples are known data of independent and dependent for Network training. Yeh et al. (1993) mentioned that the source of data set can be divided into three types: 1) a questionnaire, 2) statistical data, and 3) from the experiment. A set of data will be separated into two sets; Training set and test set. Klimasauskas (1993) commented that there result. When testing with a testing set it
Stop training Network can be performed in two ways: 1) determine training cycle (Epochs) and 2) determine an acceptable value of error. Bhokha, (1998) Carpenter (1993) suggested to determine a number of a training cycle at 20,000 - 100,000. Another method is to determine a deviation between actual data and a result that the Network can calculate (Khan et al., 1993). A caution is that training a Network too long will make a problem called over fitting as shown in Fig. 8, in which the Network can learn to achieve the least error result. When testing with a testing set it appears that it cannot give a good the result (Bhokha, 1998). Samples are known data of independent and dependent for Network training. Yeh et al. (1993) mentioned that the source of data set can be divided into three types: 1) a questionnaire, 2) statistical data, and 3) from the experiment. A set of data will be separated into two sets; Training set and test set. Klimasauskas (1993) commented that there result. When testing with a testing set it
Fig. 7 Back-Propagation Artificial Neural Network
Fig. 7 Back-Propagation Artificial Neural Network
A. Overview of Face Recognition by ANFIS Technique
A. Overview of Face Recognition by ANFIS Technique
C. Design and Development 1. Preparing a Picture for Practicing
C. Design and Development 1. Preparing a Picture for Practicing
Fig. 9 Shows a Testing Result of Efficiency to Detect Faces by ANFIS Technique The Test will be divided into two parts as follows;
Fig. 9 Shows a Testing Result of Efficiency to Detect Faces by ANFIS Technique The Test will be divided into two parts as follows;
2. Testing an Efficiency to Detect Faces y Neural Network Back Propagation Technique Fig. 10 Shows a Testing Result of Efficiency to Detect Faces by Neural Network Back Propagation Technique »bust Face Recognition Technique with ANFIS in Compared with Feedforward Backpropagation Neural Networ Using Specific Features of Wavelet Transform and HOt
2. Testing an Efficiency to Detect Faces y Neural Network Back Propagation Technique Fig. 10 Shows a Testing Result of Efficiency to Detect Faces by Neural Network Back Propagation Technique »bust Face Recognition Technique with ANFIS in Compared with Feedforward Backpropagation Neural Networ Using Specific Features of Wavelet Transform and HOt
Fig. 11 Shows a Testing Result of Algorithm Efficiency to Detect Faces by ANFIS Technique
Fig. 11 Shows a Testing Result of Algorithm Efficiency to Detect Faces by ANFIS Technique
2. Testing Algorithm Efficiency to Detect Faces by Neural Network Back Propagation Technique Fig. 14 Shows a Training Part to Test Algorithm Efficiency to Detect Faces by Neural Network Back Propagation Technique Fig. 14 Shows a Training Part to Test Algorithm Efficiency to Detect Faces by Neural Network Back Propagation Technique An average of Error Mean and Standard Deviation has a small value which results a little error. An average of Error Mean and Standard Deviation has a small value which results a little error.
2. Testing Algorithm Efficiency to Detect Faces by Neural Network Back Propagation Technique Fig. 14 Shows a Training Part to Test Algorithm Efficiency to Detect Faces by Neural Network Back Propagation Technique Fig. 14 Shows a Training Part to Test Algorithm Efficiency to Detect Faces by Neural Network Back Propagation Technique An average of Error Mean and Standard Deviation has a small value which results a little error. An average of Error Mean and Standard Deviation has a small value which results a little error.
MSE (Mean square error) implies when the value falls down continuously, how close the answer would be. From the graph shown above, it stops training at cycle 14 MSE 10e- 5x Axis is the number of y Axis which is MSE. Fig. 16 Shows Gradient, Mu, and Validation
MSE (Mean square error) implies when the value falls down continuously, how close the answer would be. From the graph shown above, it stops training at cycle 14 MSE 10e- 5x Axis is the number of y Axis which is MSE. Fig. 16 Shows Gradient, Mu, and Validation
Fig. 15 Show SMSE (Mean Square Error)
Fig. 15 Show SMSE (Mean Square Error)
Fig. 17 Shows a Graph Measurement to Measure an Efficiency of Movement Result in Value Robust Face Recognition Technique with ANFIS in Compared with Feedforward Backpropagation Neural Networl Using Specific Features of Wavelet Transform and HOC
Fig. 17 Shows a Graph Measurement to Measure an Efficiency of Movement Result in Value Robust Face Recognition Technique with ANFIS in Compared with Feedforward Backpropagation Neural Networl Using Specific Features of Wavelet Transform and HOC
Fig. 18 Shows a Graph Comparing Target and Neural
Fig. 18 Shows a Graph Comparing Target and Neural
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved