INTRODUCTION
Throughout last decades thousand works in the given area have been published, but despite it the problem of search of graphic representations in a database under their maintenance remains actual till today. It is possible to explain it complexity of the given problem caused first of all by complexity of the mechanism of visual perception of the person. The basic problem in the given area, according to many researchers, is «semantic rupture» between down-level the maintenance of the image with which the system operates, and semantics of the image necessary for the user. Also certain complexity is represented by necessity of work with great volumes multidimensional given - vectors of the signs calculated for the description of each image.
Methods of processing of images play a considerable role in researches and information systems. Search in the text description is often used, but it is insufficiently effective, as demands additional expenses of time and as subjectively depends on the person making the text summary. Newer approach to the decision of the specified problem is contextual search which is carried out on image contents according to inquiry of the user specifying visual characteristics of the required image, or the image - the sample of search.
The review of methods of search of images
The basic directions of researches by search of graphic representations
It is possible to allocate following basic directions of researches in the given area:
Allocation of signs of images. Search of various ways of the description of images and their comparison among themselves. Within the limits of the given direction all new kinds of vectors of signs and ways of their calculation, and also the new metrics set on space of these vectors are offered.
Multidimensional indexing. Working out of algorithms of the multidimensional indexing approaching for problems CBIR for which high dimension and great volumes of indexed data is characteristic.
Designing of systems of search. The important feature of any system is its ergonomics - convenience of work with it for the user. For systems CBIR this parametre plays an especial role in view of complexity of such systems. How to show to the user simultaneously a considerable quantity of images which the system has selected as the answer to inquiry? How to give to the user possibility to estimate quality of search that further was possible to consider this estimation for specification of result of search? How to construct dialogue of the user with system? Search of answers to these questions - a problem of the researchers who are engaged in designing of systems of search.
(The size of animation: 95Ęá, the Permission ŕíčěŕöčč:433x283, Quantity of repetitions: 5, Quantity of shots: 8)
(For repeated display of animation update page)
Search in the maintenance - Content Based Image Retrieval (CBIR)
Figure 1 - Traditional architecture of systems CBIR
- Allocation of signs of images
- Multidimensional indexing
- Designing of systems of search
Levels of the maintenance and search of images
By search of graphic representations in their maintenance pay attention to levels of the maintenance of the image which are in turn subdivided as follows:
Figure 2 - Levels of the maintenance of graphic representations
The text summary includes only semantics of the image. And at down-level characteristics use such factors as - the form, a structure, colour, brightness, and also their combination.
By search of graphic representations in their maintenance apply following signs:
- Colour signs
- Structure signs
- Form signs
- Spatial signs
Figure 3 - Scheme of search of graphic representations
Color
Let's characterise a sign of colour which in turn is one of signs of search of graphic representations under their maintenance.
Colour is the qualitative subjective characteristic defined on the basis of visual sensation, and depending on a number of physical, physiological and psychological factors.
The colour sign basically consists of two components:
- Histograms
- Static model
Figure 4 - Components of a sign of colour
The histogram is a way of graphic representation of tabular data.
The static model is a model of statistics which can be characterised by a number of signs: a population mean, a dispersion etc.
Stages of process of comparison of images
In the decision of a problem of contextual search of images, a method of comparison of histograms, it is possible to allocate following stages of process of comparison of two images:
- Digitization of colour space. At the given stage of colour in the initial image are replaced with colours from a demanded final set.
- Histogram construction. At this stage probabilities of occurrence of each of the colours belonging to the demanded set, in the image, and histogram normalisation pay off.
Comparison of images. Here the histograms constructed on the basis of compared images, by calculation of distances between CR the image-sample and histograms of all images from a DB are compared.
Texture
Structure - the image consisting from more or less of relatives on perception of elements. The mixed structures can include elements from several sets (classes) of elements. A structure, vicinities of which all points are similar each other name a uniform structure.
The structure is the main sign which is applied by search of graphic representations in their maintenance.
In a structure in turn allocate following signs:
Figure 8 – Textural signs
Signs Tamura
Signs Tamura are the characteristics essential to visual perception:
- Granularity (coarseness)
- Contrast (contrast)
- An orientation (directionality)
- Linearity (line-likeness)
- A regularity (regularity)
- Roughness (roughness)
The form of objects
The object form - a making sign which is applied by search of graphic representations in their maintenance.
The object form possesses a number of signs:
Figure 12 – Form signs
Requirements to form signs:
- Invariancy to parallel carrying over
- Invariancy to scale change
- Invariancy to turn
- Stability to form minor alterations
- Simplicity of calculation
- Simplicity of comparison
CONCLUSIONS
There is a wide range of various algorithms of search of graphic representations under their maintenance on each of characteristics separately:
- Colour: histograms or statistical model?
- A structure: filters Gabora, filters ICA
- The form: descriptors Furie, the invariant moments
It is possible to combine methods of search in various characteristics:
- The choice of a method of synthesis depends on a specific target (that with what
We mix)
- It is important to consider weight of sources
- The adaptive approach
Formulating precisely a search problem to reduce degree of similarity of two images to analytical expression in the form of some formal system, practically it is not possible. The reason consists in ambiguity of the concept of similarity of images. However if to allocate some experimental rules and will agree about the sizes of an error, the decision of the set problem can be defined in the form of some system of the equations.
THE LITERATURE LIST
- Askoy S., Haralick R. M. Textural features for image database retrieval. In Proc. of IEEE Workshop on Content-Based Access of Image and Video Libraries, in conjunction with CVRP’98, p. 45-49, Santa-Barbara, CA, June 1998.
- Lee J. H. Analyses of multiple evidence com¬bination. SIGIR '97: Proceedings of the 20th annual international ACM SIGIR conference on Research and development in in¬formation retrieval, New York, NY, USA: ACM Press, p. 267-276, 1997.
- Loncaric S. A survey of shape analysis techniques. Pattern Recognition, 31(8), p. 983-1001, 1998.
- Manjunath B.S., Ma W.Y. Texture features for browsing and retrieval of image data. IEEE Transactions on Pattern Analysis and Machine Intelligence, 18 (8), p. 837-842, 1996.
- Rui Y., Huang T. S., Chang S. F. Image Retrieval: Current Techniques, Promising Directions, and Open Issues. JVCIR, Vol. 10, No. 1, p. 39-62, March 1999.
- Shuang F. Shape Representation and Retrieval Using Distance Histograms. Technical Report TR 01-14, Department of Computing Science, University of Alberta, Edmonton, Alberta, Canada, October 2001.
- Stricker M., Orengo M. Similarity of Color Images. Proceedings of the SPIE Conference, Vol. 2420, p. 381-392, 1995.
- Swain M.J., Ballard D. H. Color Indexing. International Journal of Computer Vision, Vol. 7(1), p. 11-32, 1991.
|