1
1
2
Guided by, Amol Joy (Assistant professor) ECE department
Date of Presentation 20 August 2015
Presented by, Ann Mary Thomas S7 ECE A Reg.no:12010492 Roll no: 121019
FACE 2 MUSIC
3
• Introduction• Objective• How to tract facial features in images• How to tract emotions in songs• FACE 2 MUSIC• Work flow for FACE 2 MUSIC• Experimental result• Image data base• Image classification• Audio data base• Audio classification• Over all performance of FACE 2 MUSIC• Conclusion• Reference
OUTLINE OF THE PRESENTATION
4
• smart phone has introduced mobile application.
• Used in therapeutics.
• Emotion sensitive music app would be of high value.
• Image-based emotion recognition.
INTRODUCTION
5
• the entire topic is divided into two sections: part A and part B. Part A consist of ; i. How to tract facial features in images ii. How to tract emotions in songs Part B consist of; i. Workflow for Face 2 Music ii. Experimental result iii. Overall performance
contd…
6
• Introduce a mobile application .
• sense the emotions of people for their better health.
• To provide better Human-machine interactions
OBJECTIVE
7
PART A
8
1.From extracted Action Units(AUs).
2.From Motion Units(MUs).
HOW TO TRACT FACIAL FEATURES IN IMAGES ?
9
• Developed by Ekman.
• Describes a specific feature in the face.
• It achieved accuracies ranging from 64.29% to 100%.
1.From extracted Action Units(AUs)
10
• Describes the direction of motion of feature and its concentration.
• Computationally expensive and person dependent.
• Hence, AU’s would be potentially better .
2.From Motion Units(MUs)
11
• Many theories developed to assign emotions or moods to
songs.
• Some of them are, 1. Hevner’s theory 2. Russell’s axes 3. Thayer’s axes
HOW TO TRACT EMOTIONS IN SONGS ?
12
• Some include tempo, loudness and harmony
• Low level sound descriptors and several open source libraries are available.
• Here, implemented a hybrid music and theme classifier.
Contd...
13
PART B
14
• It is an android application.
• It streams music from online radio stations .
• Emotion is deduced by emotion recognition techniques .
FACE 2 MUSIC
15
WORK FLOW FOR FACE 2 MUSIC
16
FACIAL FEATURES
17
• Tested on a Samsung Galaxy Tab 10.1 model.
• Server running on a 64 bit Intel core i5 processor.
• Developed using Android 3.1 API .
• Image processing and classification code was written in Matlab R2011b.
EXPERIMENTAL RESULTS
18
• Testing results of the audio and image classifiers are presented separately.
A. Image data base B. Image classification C. Audio database D. Audio classification E. Overall Face 2 Music performance
Contd...
19
• Image classifiers were trained on a home-made data base .
• Data base is composed of 120 images of 8 people.
• 5 images for each acted emotion per person.
• Images were classified in to 3 classes, i. Neutral ii. Sad iii. Happy.
1. IMAGE DATA BASE
20
SAMPLE IMAGES FROM THE DATA BASE
21
• Trained and tested on neural network and SVM classifiers.
• A grid search was performed to find the best classifier .
• Majority vote classifiers, given different feature sets.
• It is composed of 1 vs.1 SVM classifiers using RBF kernel.
2. IMAGE CLASSIFICATION
22
• Most of the misclassified images were from the sad class.
• Best classifiers were saved and used to classify the images .
• The feature set containing distances produced best results.
Contd...
23
24
• Classifier architectures were trained and tested on the Million song dataset subset.
• It included 10,000 instances and 4 features.
• Instances were clustered using k-means clustering algorithms.
3. AUDIO DATA BASE
25
• A neural network and an LM-SVM classifiers were trained.
• A grid search was performed to find optimal box constraint and kernel parameters.
• Training was validated using 5-fold, cross fold validation.
• Best models of various classifiers were saved on android tablet.
4. AUDIO CLASSIFICATION
26
5. OVERALL FACE 2 MUS PERFORMANCE
• The size of the application is 272KB.
• Uses 12 to 24 MB of RAM .
• Start up time is approximately 2 minutes.
• Time can vary based on the internet connection.
• To reduce start up time, run audio processing blocks on server.
27
28
• Proposed an automated online song streaming application .
• Emotions detected using facial feature identification, tracking and classification .
• Accuracy can be enhanced by more robust features.
• Real time performance can be improved.
CONCLUSION
29
• M. Walsh (2011, October 7). Mobile App Biz Soars: $12B by 2015.
• The Independent (2011, November 9). Mobile application
trends for 2012: the top ten applications. • J. Imam (2012, June 16). Young listeners opting to stream, not
own music.
• C.E. Guzzetta. “Effects of relaxation and music therapy on patients in a coronary care unit with presumptive acute myocardial infraction.” Heart & Lung: The journal of critical
care, vol.18, issue 6, p.609, 1989.
REFERENCES
THANK YOU
31