Presentation is loading. Please wait.

Presentation is loading. Please wait.

Zhengjun Pan and Hamid Bolouri Department of Computer Science

Similar presentations


Presentation on theme: "Zhengjun Pan and Hamid Bolouri Department of Computer Science"— Presentation transcript:

1 High Speed Face Recognition Based on Discrete Cosine Transforms and Neural Networks
Zhengjun Pan and Hamid Bolouri Department of Computer Science University of Hertfordshire Presented By Mustafa Mirac KOCATÜRK High Speed Face Recognition Based on DCT and Neural Networks

2 High Speed Face Recognition Based on DCT and Neural Networks
OUTLINE Introduction to the Face Recognition Existing Methods for Feature Extraction and Advantages Using DCT Key Characteristics of Recognition Systems Information Packing Using DCT System Description of DCT Recognition System Brief Information about ORL Database Experimental Simulations Conclusion High Speed Face Recognition Based on DCT and Neural Networks

3 High Speed Face Recognition Based on DCT and Neural Networks
INTRODUCTION Face recognition is the science of programming a computer to recognize a human face. The steps of Face Recognition are Face Detection (Feature extraction) Face Normalization Face Identification High Speed Face Recognition Based on DCT and Neural Networks

4 High Speed Face Recognition Based on DCT and Neural Networks
INTRODUCTION The Key Characteristics of the Recognition Systems are: Recognition Rate Training Time Recognition Time High Speed Face Recognition Based on DCT and Neural Networks

5 High Speed Face Recognition Based on DCT and Neural Networks
INTRODUCTION Existing Computational Models For Feature Extraction: Geometrical Features Statistical Features Feature Points Neural Networks High Speed Face Recognition Based on DCT and Neural Networks

6 High Speed Face Recognition Based on DCT and Neural Networks
INTRODUCTION Problems of Existing Systems are: High Information Redundancy Building a Database of Faces Computationally Expensive Spare Computation Time for Real-Time Applications High Speed Face Recognition Based on DCT and Neural Networks

7 High Speed Face Recognition Based on DCT and Neural Networks
INTRODUCTION The Advantages of DCT: Removes the redundant info Decreases the computational complexity Much faster than the other models High Speed Face Recognition Based on DCT and Neural Networks

8 DISCRETE COSINE TRANSFORM
High Speed Face Recognition Based on DCT and Neural Networks

9 DISCRETE COSINE TRANSFORM
DCT is being used as a standard in JPEG files High Speed Face Recognition Based on DCT and Neural Networks

10 DISCRETE COSINE TRANSFORM
How many coeffiecents should be used? High Speed Face Recognition Based on DCT and Neural Networks

11 DISCRETE COSINE TRANSFORM (coefficient analysis)
High Speed Face Recognition Based on DCT and Neural Networks

12 DISCRETE COSINE TRANSFORM (coefficient analysis cont.)
High Speed Face Recognition Based on DCT and Neural Networks

13 DISCRETE COSINE TRANSFORM (subimage analysis)
High Speed Face Recognition Based on DCT and Neural Networks

14 DISCRETE COSINE TRANSFORM (subimage analysis cont.)
High Speed Face Recognition Based on DCT and Neural Networks

15 High Speed Face Recognition Based on DCT and Neural Networks
SYSTEM DESCRIPTION The main idea is to apply the DCT to reduce information redundancy and use the packed information for classification System consists of Coefficient Selection Data Representation High Speed Face Recognition Based on DCT and Neural Networks

16 High Speed Face Recognition Based on DCT and Neural Networks
ORL DATABASE Built at Olivetti Research Laboratory 400 images 10 for each 40 distinct objects 4 female and 36 male subjects 92 X 112 pixels each with 256 gray level Images differ in; Lightning Facial expressions Facial Details High Speed Face Recognition Based on DCT and Neural Networks

17 SIMULATIONS OF DCT (experimental setup)
MLP are initialised to random values [-0.5,0.5] Learning Parameters set to 0.02,0.008,0.0001 The max. number of training epochs is 1000 The multiplication factor of β is set to 1.1 Training samples are randomed to avoid the influence of the presentation order 200 training and test images are used (First 5 of the each 40 outputs are for training and testing) High Speed Face Recognition Based on DCT and Neural Networks

18 SIMULATIONS OF DCT (experimental setup cont.)
T-Tests are based on the 0.05 level of significance T-Test statistics has to exceed for experimental results to be classified as statistically different from the reference case. The reference case of the system is 35 DCT Coefficents 75 Hidden Neurons High Speed Face Recognition Based on DCT and Neural Networks

19 SIMULATIONS OF DCT (# of coefficients)
High Speed Face Recognition Based on DCT and Neural Networks

20 SIMULATIONS OF DCT (# of hidden neurons)
High Speed Face Recognition Based on DCT and Neural Networks

21 SIMULATIONS OF DCT (sub-image size)
High Speed Face Recognition Based on DCT and Neural Networks

22 SIMULATIONS OF DCT (different recognition approaches)
High Speed Face Recognition Based on DCT and Neural Networks

23 High Speed Face Recognition Based on DCT and Neural Networks
CONCLUSION DCT using Neural Networks is a very fast and efficient approach in face recognition. Truncating the unnecessary info reduces computational complexity. The experiments reported above demonstrate that using only %0.34 of the DCT coefficients produces a respectable recognition rate while the processing time is 2 times faster. High Speed Face Recognition Based on DCT and Neural Networks


Download ppt "Zhengjun Pan and Hamid Bolouri Department of Computer Science"

Similar presentations


Ads by Google