A comparative study of ability of the proposed novel image retrieval algorithms is performed to provide automated object classification invariant of rotation, translation, and scaling. Simple cosine similarity coefficient methods are analyzed. Considering applied cosine similarity coefficient methods, the two following approaches were tested and compared: the processing of the whole image and the processing of the image that contains edges extracted by the application of the Sobel edge detector. Numerical experiments on a real database sets indicate feasibility of the presented approach as an automated object classification tool without special image pre-processing.
Top2. Proposed Methods For Automatic Object Classification
Three different methods are proposed in this chapter invariant of rotation, translation or scaling of the classified objects. They successfully perform object classification on set of three different groups of objects Dinosaurs, Mummies and Sculls represented by images taken under various rotational, scaling and zooming conditions.
Two techniques for automatic object classification are applied. Sobel edge filtered images are used for similarity computation in the first method and in the second method simple cosine similarity coefficient is applied on plain gray images with the goal to classify them.
The first technique implies procedure with an image converted to gray image with the extracted edges using Sobel edge detection method (Engel, 2006), (Jähne, 1999), (Farid, 2004), (Kroon, 2009), (Scharr, 2007), (Gonzalez, 2001). The idea behind this method is to significantly reduce the amount of data and filter out useless information, while preserving the important structural properties of an image and the targeted object.
Every image is processed as a two-dimensional m×n matrix image. The two-dimensional Sobel masks are applied to gray images. The Sobel operator performs a 2-D spatial gradient measurement on an image. It is used to find the approximate absolute gradient magnitude at each point in an input grayscale image. The Sobel edge detector uses a pair of 3×3 convolution masks, one estimating the gradient in the x-direction (columns) and the other estimating the gradient in the y-direction (rows). After that the magnitude of the gradient is calculated. In the next step cosine similarity coefficient (Singhal), (Garcia, 2005), (Tan, 2005), (Zeljković, 2007) is applied in order to extract the image containing the most similar object in the database.