Applied Mathematics & Information Sciences

Author Country (or Countries)



In a Segmentation-based approach, an image is segmented and its various regions are classified, unlike classifying the individual pixels. This papers uses the ReNet architecture to extract the features of an object in an image. This ReNet architecture replaces each convolutional layer(CNN) with four RNNs that also brings together lower-layer features from different directions. After the extraction of feature the image is over segmented into superpixels first and then it is classified into individual superpixels. The dependencies to the nearby superpixel labels shall be explored and exploited by Conditional Random Field statistical approach. Though the time to segment and label the images is somewhat higher, the pixel accuracy is more when this technique is implemented in the two datasets SIFT Flow and Stanford Background Dataset.

Digital Object Identifier (DOI)