Humans make object recognition look inconsequential. In this chapter, scale-invariant feature extraction and shape-index depiction are used on a range of images for identifying objects. The shape-index is attained and used as a local descriptor or key-point descriptor. First surface properties for shape index identification and second as 2D scale invariant feature transformed for key-point detection and feature extraction. The object recognition classification is compared results with shape-index identification and 2D scale-invariant feature transform for key-point detection with SIFT and SURF. The authors are using images from the ImageNet dataset, and with use of shift-index + SIFT descriptors, they are finding better accuracy at the classification stage.
Top1. Introduction
Object recognition is an inspiring task because of its unpredictable features for diverse objects. This application is implemented by taking an image as an input and from that an object is identified using the dataset which encompasses trained images. Different algorithms are proposed by some researchers (Lowe, 2004; Chowdhary, 2011; Chowdhary, Muatjitjeja & Jat, 2015) to make this recognition easy. In contemporary surveillance classification, scanning an item/object from an apprehended image from a camera became a very tricky job for all the surveillance professionals. Those professional want an expert system to recognize the objects. Recognizing objects from a rotated, scale invariant and occluded images is a challenging task and hard to identify an object from 2D and 3D view images. But the issue is the results from those expert systems is not that much exact as we are predicting. So, such a system is required which produces more precise results. Identifying object from an image with more noise and occlusion is also a confusing task.
The standing methodologies recognize an object by its size, structure and position. Such systems may be failed to find an object which is in upside down or a slight change in its position. This failure causes due to the reason that it is compared with its default position. It is required to develop such an expert system to recognize and understand the 3D-structure of an object. After adopting 3D-structures of an object for object recognition, the results are not failed in the cases where the system tends to detect the object in every position. High-definition (HD) cameras are like human eyes in recent days. HD cameras are highly trending technology in present world because they capture images with more accurate pixels, 3D sensing and displaying. If the cameras are auspicious to capture accurate pixels and equally important 2D, 3D sensing, capturing and displaying then it is possible to identify and recognize the objects which has been captured by those high-definition cameras. It is found difficult to recognize or identify the objects on the surveillance system with normal cameras so the high-definition cameras are good option to capture 2D and 3D view images which lead to design a system to identify and recognize an object which may assist surveillance experts. Usual surveillance systems available in airports, temples, malls and different public places are using object recognition with low accurate results so it is needed to to develop an expert system with more accurate results to identify and recognize an object.
In most of the object recognition system find it difficult in recognition when the object is occluded by another object. Some object is taken and segmented, then stored as dataset which is mainly used in map. Using this dataset the scanned image is identified. The identified object is depending on the feature generated from the scanned image and that feature point is mapped onto the object using the dataset. Objects in a captured or sensed image had been identified and recognized by the expert system. Even now also there are numerous expert systems helping for surveillance experts for security purposes but the issue is how accurate the expert system is producing the results. In order to increase the accuracy, we are proposing an expert system which includes two mechanisms to recognize an object from a captured image. This mechanism can be applied to any range image where an image is with scale invariant, rotation and occluded images.
The shape-index is defined as a single valued measure of local curvature which is derived from the eigenvalues of the Hessian. The shape-index can be castoff to invention erections constructed on their apparent local shape. The shape index maps to values from -1 to 1 which signify diverse type of shapes (Koenderink & van Doorn, 1992). A shape index is actually of spatial measurement. This measurement marks the numerical connection among mathematical modelling and empirical study (Prasad, Gupta & Biswas, 2001).