Technical Field
[0001] The embodiments herein generally relate to a method for local binary pattern based image retrieval system using compressive sensing.
Description of the Related Art
[0002] While many computer applications are used both on personal computers and networked systems, the field of information retrieval and database access for casual users has garnered considerable interest. With rapid grow in the size of images database on the repositories , text based searching of images has the issue of accuracy as the database of images are stored in unstructured format on the repository, leading to a need for search engines and access portals based on content of the image or videos stored.
[0001] Search Engine are the most popular software now a days to look for any information on the internet. For searching of any information, text based searching is popular but to search any image of our interest on the internet we have to describe it in words about the image, it may be name of object or place in it or any activity, means we have to describe the characteristic of that image into words to search it on the internet and similarly at the time of storing that image on the repositories, its description is to be also saved called metadata. Correct annotation is the major issue in accuracy of searching of images on the internet. This issue has been addressed by developing reverse image search engine where instead of Meta data directly user gives the input in the form of a sample image and look for the similar images from the repositories.
3
[0002] Content based image retrieval systems have been developed from last three decades but still very few commercial reverse image search engines are available like recently introduced Google Lens. There are two open research problems still exist in the content based retrieval systems, first is the semantic gap and second is the curse of dimensionality.
[0003] Sematic gap refers to the gap between the description given by the user and the description perceived by the search engine. The objects presents in an image have different descriptions for the different persons, it is difficult for the computer system to understand the perceptions of all the users for same image query.
[0004] Second problem in content based image retrieval system is curse of dimensionality. To improve the performance of the content based retrieval system multi feature based fusion is performed to form the visual descriptor. The size of the visual feature descriptors gets increased with this multi feature fusion. For large feature size representation of the image degrades the performance of the distance matching methods. From the foregoing discussion, it should be apparent that a need exists for a method that can represent the properties of image in compact form so that performance can be improved as well as the fast retrieval is possible.
4
SUMMARY
[0005] In view of the foregoing, an embodiment herein provides a method for implementation of content based image retrieval system that can provide better results.
[0006] Local Binary pattern based methods are popular in various applications of digital image processing including segmentation, object recognition, image compression and classification applications. The advantage of Local binary pattern generated for an image are scale invariant, not affected by lighting conditions , so this method is suitable for designing content based image retrieval system.
[0007] Local binary pattern and its improved methods like ternary and tetra patterns are also used in designing of content based retrieval systems. This method generates 59 feature vector for an image.
[0008] Compressive sensing is an approach to signal acquisition and processing that makes use of the inherent properties of some signals to measure and mathematically reconstruct the signal based on a limited series of test measurements. This disclosure relates to novel systems and methods for use of sensing matrix of compressive sensing theory. This theory suggest that if the signal is spares in nature and follows the restrictive isometry property and columns of the matrix are incoherence , then perfect signal recovery is possible with very few samples.
[0009] There are two steps in compressive sensing for image reconstruction, first is the sampling of the image. Natural images are considered as sparse in nature having very few non zero values or mostly near to zero, if can be represented in suitable transform basis becomes sparse and then very few measurement vectors can be calucated by multiplying the trafomed image matrix with the sensing matrix. Existing literature has very few studies on designing
5
Content based image retrieval system suing compressed sensing concepts. Motivated by advantages of compressed sensing in compact representation of the image visual features fast retrieval systems can be designed.
[0010] These and other aspects of the embodiments herein will be better appreciated and understood when considered in conjunction with the following description and the accompanying drawings. It should be understood, however, that the following descriptions, while indicating preferred embodiments and numerous specific details thereof, are given by way of illustration and not of limitation. Many changes and modifications may be made within the scope of the embodiments herein without departing from the spirit thereof, and the embodiments herein include all such modifications.
BRIEF DESCRIPTION OF THE DRAWINGS
[0011] The embodiments herein will be better understood from the following detailed description with reference to the drawings, in which:
[0012] FIG. 1 illustrates a method for implementation of content based retrieval system using compressed sensing theory concepts based on local binary pattern according to an an embodiment herein.
DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS
[0013] The embodiments herein and the various features and advantageous details thereof are explained more fully with reference to the non-limiting embodiments that are illustrated in the accompanying drawings and detailed in the following description. Descriptions of well-known components and processing techniques are omitted so as to not unnecessarily obscure the embodiments herein. The examples used herein are intended merely to facilitate an understanding of ways in which the embodiments herein may be practiced and
6
to further enable those of skill in the art to practice the embodiments herein. Accordingly, the examples should not be construed as limiting the scope of the embodiments herein.
[0014] FIG. 1 illustrates a method for implementation of content based retrieval system using compressed sensing theory concepts based according to an embodiment herein.
[0015] Content-based image retrieval (CBIR) is a powerful concept for finding images based on image contents, and content-based image search and browsing have been tested using many CBIR systems.
[0016] The invention provides a content based image retrieval system comprising: a feature image database extracted through local binary pattern based method. A useful texture descriptor for images is the Local Binary Pattern (LBP), which thresholds adjacent pixels based on the value of the current pixel. The local spatial patterns and the contrast in the grey scale in an image are effectively captured by LBP descriptors. For an example of a finger image which can be used to develop a biometric systems, it is crucial to employ the ridge-valley intensity contrast in order to trace the ridge lines. In comparison to the valleys in their immediate surroundings, the edge lines on the ridges are more intense. LBP based method efficiently traces the ridge lines in a finger image by embedding the spatial structure into its descriptor. LBP is frequently applied in a variety of computer vision applications
[0017] The four-step technique for computing the LBP descriptor from a picture is described below. For each pixel element (m, n) in an image, X, choose Y neighboring pixels at a radius T.
[0018] Determine the intensity difference between the current pixel (m, n) and its K adjacent neighbors.
[0019] Now in the next step is thresolding is performed to make zero all negative
7
difference values to zero and all positive value to be made 1.
[0020] Replace the intensity value at (m, n) with the equivalent decimal value obtained from the K-bit vector.
[0021] In the offline mode training images are given to the system to generate the Local binary pattern
[0022] After getting the local binary pattern for image compressive sensing is performed.
[0023] Compressive sensing (also known as compressed sensing, compressive sampling, or sparse sampling) is a family of signal acquisition and processing techniques for efficiently acquiring and reconstructing a signal. As used herein, the term “signal” and its grammatical equivalents includes, but is not limited to, intensity, frequency, or phase data as it pertains to an electrical, electromagnetic, or magnetic field, as well as to optical or non-optical image data, spectral data, diffraction data, and the like. In compressive sensing, reconstruction of a signal is performed by making a limited number of signal measurements according to a defined set of sampling functions (or test functions), and subsequently finding mathematical solutions to the resulting system of linear equations that relate the unknown “true” signal to the set of measured values. Reconstruction thus provides an estimate of the “true” signal, the accuracy of which is dependent on several factors including, but not limited to, properties of the signal itself, the choice of test functions used to sample the signal, the amount of noise in the signal, and the mathematical algorithm selected to solve the system of linear equations. Because the signal is under-sampled, the system of linear equations is underdetermined. In general, underdetermined systems of equations have an infinite number of solutions. The compressive sensing approach is based on
8
the principle that prior knowledge of or reasonable assumptions about the properties of the signal can be exploited to recover it from far fewer sampling measurements than would be required by conventional Nyquist-Shannon sampling. [0024] In favorable cases, variants of the standard algorithms described in the compressive sensing literature can be used to reconstruct tens or even hundreds of reconstructed frames of video data from a single such data acquisition period. This type of compressive sensing system has been demonstrated for optical video cameras, and researchers are currently attempting to apply the same approach to compressive sensing in transmission electron microscopes. The temporal compressive sensing method described above that is potentially applicable to a wide variety of signal acquisition and processing fields in addition to optical video and electron microscopy. In addition, several distinct hardware implementations of the approach are disclosed that enable operation in very different time domains.
[0025] The derived local binary Patterns are now reduced using Gaussian random sensing matrix. First the random sensing matrix is generated of size 45x59.
[0026] The local binary pattern for each image has 59 feature vectors, to reduce the dimensions of the of the feature space random sensing matrix is multiplied
[0027] The resultant measurement vectors will be of 45 in size for each image in the database.
[0028] For the query image same process is followed and then similarity matching is performed using Euclidian distance measure. In accordance with one or more embodiments, the compressive sensor produces the better precision and recall with lesser number of feature vectors and hence reduces the retrieval time, make method fast and also reduces the storage size requirements of the database.

I/We Claim:

1.A method for local binary pattern-based retrieval system using compressive sensing, wherein 1 the methods comprise: 2
a novel framework is proposed for content based image retrieval system; 3
local Binary patterns are applied on the image to get the feature descriptor; 4
a novel application of compressed sensing concept to reduce the dimensions of the 5 features.