Facial Composite Sketch Matching to Images Applying Convolutional Neural Network

This research paper aimed to test the performance of developed facial matching composite sketch by focusing on image processing using convolutional neural network (CNN) and compared its accuracy with the Forensic Sketch Matching to Mug shots using Scale-Invariant Feature Transform. This study will help law enforcement officials in finding suspected criminals with forensic sketches through the available mug shots. Previous studies shown that using CNN models for visible spectrum face recognition improved the heterogeneous face recognition. The researchers used experimental approach as its methodology to test the performance of the developed facial recognition application in terms of its precision, recall and f-measure. Data sets of 75 frontal view images and its corresponding facial composite sketch was feed and processed in the developed facial matching application. T-test was used to test the significant difference of the two application. The results of performance are as follows: precision - 86%; recall - 89%; and f1-score - 87% and its accuracy is 89%. The computed t value 4.9206 which is higher than the t critical value of 1.9925 means it failed to accept the null hypothesis. Therefore, there is a significant difference between the accuracy of the two applications. This study proved that CNN has a better performance in matching sketch images to its mugs shots.