Region depth ordering in monocular single image using statistical blurring measure

Abstract

This paper proposed a region depth ordering method from monocular single image by using multi-layer back-propagation neural network. The feature that used in this work is the wavelet coefficients of the training images. The feature extraction results are the different of the object’s texture information of the areas in the images. The learning results are the relationship between the logical depth of the object and the different of area’s texture information in the training images. In this paper, we used 125 images to be the training images by divided into 5 groups for validation. Experimental results show that performance in object’s depth ordering that the correctness is more than 94% on average.