Page 60 - Flytxt
P. 60
for encoding. Since training data could quality on the fifteen scene categories
have several images, total number of data set. The average accuracy of the
DAISY descriptors could be very large (in model was 76.4% in the case of a 40%–
millions). To avoid complexities during 60% random split of images into training
clustering and facilitate fast encoding, we and testing data sets respectively
use Mini-Batch K-Means algorithm. The
histogram corresponding to each image
is augmented with its HOG descriptor Through artificial intelligence,
machines have come closer
using a pooling procedure, to generate to human ability in several
the final feature vector for different cognitive tasks such as
images. The associated class label (e.g. visualizing and identifying
living room, store etc.) would be readily diverse objects and
available since the training data set is environments. Consequently,
pre-labelled. deep learning has emerged
A multi-class SVM (each class as a powerful tool to solve
corresponds to a scene category such problems involving machine
as living room, store etc.) is trained and vision and perception.
cross validated to assess the model
60 INSIGHTZ - VOLUME 03, 2018
