Content-based image retrieval based on emergence index

Content-based image retrieval based on emergence index

Emergence is a phenomenon where we study the implicit or hidden meaning of an image. We introduce this concept in image database access and retrieval of images using his as an index for retrieval. This would give an entirely different search outcome than ordinary search where emergence is not considered, as consideration of hidden meanings could change the index of search. A feature of an image, which is not explicit would be emergent feature if it can be made explicit. There are three types of emergence: computational emergence, thermodynamic emergence and emergence relative to a model. In computational emergence, it is assumed computational interactions can generate different features or behaviors. This is one of the approaches in the field of artificial life. Thermodynamic emergence is of the view that new stable features or behaviors can arise from equilibrium through the use of thermodynamic theory. In emergence relative to a model, deviation of the behavior from the original model gives rise to emergence. We would use this latter view in our work. Two classes of shape emergence have been identified: embedded shape emergence and illusory shape emergence. In embedded shape emergence all the emergent shapes can be identified by set theory procedures on the original shape under consideration. For example, in a set S= {a,b,c,d,e}, we can find subsets like S1={a,b,c}, S2={c,d,e}, S3={a,c,e} and so on. But in illusory shape emergence, where contours defining a shape are perceived even though no contours are physically present, this kind of set theory procedures are not enough and more effective procedures have to be applied to find these hidden shapes. These procedures could be based on geometrical, topological or dimensional studies of the original shape. Content-based Image Retrieval (CBIR) techniques, so far developed, concentrated on only explicit meanings of an image. But more meanings could be extracted when we consider the implicit meanings of the same image. To find out the implicit meanings, we first destroy the shape of the original image which gives rise to unstructured image. Then we process the unstructured image to bring out the new emergent image. We discuss emergence, calculation of emergence index and accessing multimedia databases using emergence index in this dissertation. To calculate emergence index in the access of multimedia databases, we take an input image and study the emergence phenomenon of it. Also we study the emergence phenomenon of the images of the database. Both input image and images of database would give rise to more meanings because of emergence as we explained earlier. Based on the new meanings, wherever there would be a match between input image and images of database, we would pick that record up for selection. We defined emergence index as EI = f(D,F,V,C,E) where D stands for domain of the database, F for features of the image, V for various variables that define the image, C for constraints which represent the image and E for emergence phenomenon. We calculate these five variables to get emergence index for each image of the database. Also we calculate these five variables for input image as well. We talk about global aspects of features. It means features of the entire image. Examples are area, perimeter or rectangles, triangles. In some searches, to consider the global features could be advantageous in that a symmetry with the input image could be obtained on the basis of global features only. But as is clearly the case, to consider global features could overlook the individual objects that constitute the image as a whole. In the kind of searches we propose, we take into account the global features of the image of the database while considering in detail local features. Various objects that lie within an image constitute local features. In our example, there are three objects in the image, namely, a lake and two houses. Studying the features of these three objects would add to studying the features of the image globally. We took the example of a geographic location in the thesis and then showed how destruction of original image is done and further processing of the unstructured image gives new emergent image. Partial implementation of this concept is also presented at the end. In implementation, we consider the retrieval of image globally. We do not consider break-up of image into multiple objects which is left for future research.