Event Details

Enhancing Real-time Multi-class Image Classification for Binary Images Using Hardware

Presenter: Narges Attarmoghaddam
Supervisor:

Date: Wed, August 11, 2021
Time: 11:00:00 - 00:00:00
Place: ZOOM - Please see below.

ABSTRACT

Place: REMOTE Via Zoom

Zoom Meeting

https://uvic.zoom.us/j/88684184009?pwd=bWVGMzh1aCttakVrc2NMWU1KTmZKUT09


Dial by your location
        +1 778 907 2071 Canada
        +1 647 558 0588 Canada


Meeting ID: 886 8418 4009
Password: 271287


Find your local number: https://uvic.zoom.us/u/kbgw8xnjYd

Note: Please log in to Zoom via SSO and your UVic Netlink ID

 

Abstract:

Over the past decade, image classification has been an active research area in computer vision due to its many applications, such as intelligent robots, smart homes, monitoring, and surveillance. Developing image classification modules on embedded systems is one of the most complex tasks since only limited resources are available in embedded devices. At the same time, additional demanding constraints such as reliable classification accuracy, high-throughput performance, power-efficient, and real-time computing are required to be fulfilled. Due to the potential for parallelism, low power consumption, scalable resource utilization, and reconfigurability, Field Programmable Gate Array (FPGA) devices are well suited to overcome the system implementation challenges.
In this work, an area-efficient multi-class image classification system using Histogram of Oriented Gradients (HOG) feature extractor and Support Vector Machine (SVM) classifier is proposed based on binary images. In our proposed system, HOG features are extracted from binary images to simplify the feature extraction process. In addition, block normalization of the HOG is replaced with binarization to reduce hardware resource utilization. We were able to speed up the classification process compared to a similar existing solution,  while utilizing fewer hardware resources. Moreover, the proposed compact multi-class classification system shows an 11.4% higher classification accuracy than previous work using the same test setting.