Title:A Novel Decision Support for Composite Sketch Matching using Fusion of Probabilistic Neural Network and Dictionary Matching
VOLUME: 13 ISSUE: 2
Author(s):Steven L. Fernandes* and Josemin G. Bala
Affiliation:Research Scholar, Electronics & Communication, Karunya University, Coimbatore, Tamil Nadu, Professor, Electronics & Communication, Karunya University, Coimbatore, Tamil Nadu
Keywords:Local binary pattern, probabilistic neural network, dictionary matching, SketchCop FACETTE, dempster-shafer, Proportional
Conflict Redistribution.
Abstract:Traditional pencil sketches drawn by skilled pencil sketch artists are a product of illustration
based on exaggeration; there is always some amount of discrepancy between the description of
eye-witnesses and depiction of the offender by the sketch artist. To overcome this difficulty, law
enforcement agencies worldwide have started using composite sketches (sketches created using
computer). Composite sketches have obviated the need of a skilled sketch artist. Composite sketches
can be easily drawn by eyewitnesses using face design system software (SketchCop FACETTE)
in a very short time period without any prior specialized software training. Matching composite
sketches with photos available in database are still a challenging task. In this paper, a novel technique
is proposed to match composite sketches with photos available in database. The key contribution
of the proposed system in this paper is an efficient and novel methodology developed to match
composite sketches with photos available in databases which could be used by law enforcement
agencies. The photos taken from law enforcement agencies databases are passed to face detection
module. On the detected faces and composite sketches, feature extraction and classification are performed
using Multi-resolution uniform Local Binary Pattern (LBP) and Probabilistic Neural Network
(PNN) to generate score 1. Similarly, on detected faces and composite sketches, feature extraction
and classification are again performed using Dictionary Matching (DM) to generate score
2. The generated scores are collected using Dempster-Shafer (DS) theory and Proportional Conflict
Redistribution rule no. 5 (PCR5). In this study, the authors have performed pilot testing of their
technique and results of their analysis are presented to the readers.
