Electrical and Computer Engineering ETDs

Publication Date

Summer 7-13-2019

Abstract

Convolutional Neural Networks (CNN) have provided new and accurate methods for processing digital images and videos. Yet, training CNNs is extremely demanding in terms of computational resources. Also, for simple applications, the standard use of transfer learning also tends to require far more resources than what may be needed. Furthermore, the final systems tend to operate as black boxes that are difficult to interpret.

The current thesis considers the problem of detecting faces from the AOLME video dataset. The AOLME dataset consists of a large video collection of group interactions that are recorded in unconstrained classroom environments. For the thesis, still image frames were extracted at every minute from 18 24-minute videos. Then, each video frame was divided into 9x5 blocks with 50x50 pixels each. For each of the 19440 blocks, the percentage of face pixels was set as ground truth. Face detection was then defined as a regression problem for determining the face pixel percentage for each block. For testing different methods, 12 videos were used for training and validation. The remaining 6 videos were used for testing.

The thesis examines the impact of using the instantaneous phase for the AOLME block-based face detection application. For comparison, the thesis compares the use of the Frequency modulation image based on the instantaneous phase, the use of the instantaneous amplitude, and the original gray scale image. To generate the FM and AM inputs, the thesis uses dominant component analysis that aims to decrease the training overhead while maintaining interpretability.

The results indicate that the use of the FM image yielded about the same performance as the MobileNet V2 architecture (AUC of 0.78 vs 0.79), with vastly reduced training times. Training was 7x faster for an Intel Xeon with a GTX 1080 based desktop and 11x faster on a laptop with Intel i5 with a GTX 1050. Furthermore, the proposed architecture trains 123x less parameters than what is needed for MobileNet V2. The FM-based neural network

architecture uses a single convolutional layer. In comparison, the full LeNet-5 on the same image block using the original image could not be trained for face detection (AUC of 0.5).

Keywords

AM-FM, Phase, CNN, Face, Dectection

Document Type

Thesis

Language

English

Degree Name

Computer Engineering

Level of Degree

Masters

Department Name

Electrical and Computer Engineering

First Committee Member (Chair)

Marios Pattichis

Second Committee Member

Ramiro Jordan

Third Committee Member

Balasubramaniam Santhanam

Third Advisor

Marios Pattichis

Fourth Committee Member

Victor Murray

Fifth Committee Member

Andreas Panayides

Download

Share

COinS