Evaluation of Image Segmentation and Object Recognition Algorithms for Image Parsing

Abstract

The goal of this effort is to implement several algorithms for image segmentation and object recognition, unify the algorithms, and determine which approach works the best based on certain measures. Based on the number of segments produced by the segmentation implementations, there is over-segmentation or incorrect segmentation (according to a human s perception). The performance of the segmentation could have influenced the results of the object recognition. The results for precision, recall, and F-measure indicate that the best approach to use for image segmentation is Sobel edge detection and to use Canny or Sobel for object recognition. The process for this report would not work for a warfighter or analyst. It has poor performance. Additionally, its lack of variety among the algorithms reduces the chance of correctly labeling the objects in an image.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Sep 01, 2013
Accession Number
ADA587935

Entities

People

  • Amanda Lannie
  • Michael J. Wessing

Organizations

  • Air Force Research Laboratory

Tags

Communities of Interest

  • Human Systems

DTIC Thesaurus Topics

  • Air Force
  • Air Force Research Laboratories
  • Algorithms
  • Change Detection
  • Computations
  • Computer Vision
  • Data Sets
  • Detection
  • Detectors
  • Identification
  • Image Recognition
  • Image Segmentation
  • Object Recognition
  • Precision
  • Recognition
  • Test Sets
  • Training

Fields of Study

  • Computer science

Readers

  • Computer Vision.
  • Defense Acquisition Program Management