MINING IMAGE DATASETS
ABSTRACT
Mining Image dataset is one of the necessary features in the present development. The image datasets are used to store and retrieve the precious information from images, data mining on them. Pixel-wised image features were extracted and transformed into a database-like table which allows various data mining algorithms to make explorations on it. Each tuple of the transformed table has a feature descriptor formed by a set of features in combination with the target label of a particular pixel. With the label feature, we can accept the decision tree algorithm to understand relationships between attributes and the target label from image pixels, and to make a model for pixel-wised image processing according to a given training image dataset. The model can be very efficient and helpful for image processing and image mining It is likely that by using the model, various existing data mining and image processing methods could be worked on mutually in different behavior. Our model can also be used to create new image processing methodologies, refine existing image processing methods, or act as a powerful image filter.
WHAT IS PROPOSED MODEL?
An image mining method that works at a higher generality level for mining image associations is proposed. In contrast to that, our model works on a relative low generality level for image pixel classification. Pixel-wise image classification is an essential part of many image segmentation methods, for example, determining pixels of an edge (corner) in edge (corner) detection methods, pixels of a particular object in objects segmentation based methods, pixels of abnormal tissue of medical image processing, and pixel classes in thresholding, etc. The model can be usedto mine hidden relationships between an image’s pixel and its class label, and determine the interrelated features. Besides, the created model can be applied to perform pixel-wise segmentation on input images.
These two phases are: image transformation and image mining.
(1) Image Transformation Phase: This relates to how to transform input images into database-like tables and encode the related features.
(2) Image Mining Phase: This relates to how to apply data mining algorithms on the transformed table and find useful information from it. It is remarkable that the segmentation model is efficient, and requires only one scan of the date set. It can be used to effectively solve the time-consuming problem of segmentation with networks. Here we suggest two manners to apply our approach in similar situations.
The first one is using our model to substitute the existing method with the strategies mentioned above. The second one is using our model to quickly filter out the images that need advanced examinations. For example, after singling out suspicious mammograms that might contain pixels of cancer, one can apply the original method for second segmentation. The first manner is suitable for the case that our segmentation method result is better than the original one or the loss of correctness does not make significant difference. The second one is suitable for the case that the segmentation result is used in a critical manner, and the model is unable to reach that Requirement level. The model can easily extend from 2D to 3D image processing without making a revolution and the created model can generate very efficient and compact code.
IMAGE MINING PHASE
Image mining deals with the extraction of image patterns from a large collection of images. Clearly, image mining is different from low-level computer vision and image processing techniques because the focus of image mining is in extraction of patterns from large collection of images, whereas the focus of computer vision and image processing techniques is in understanding and/or extracting specific features from a single image. While there seems to be some overlaps between image mining and content based retrieval (both are dealing with large collection of images), image mining goes beyond the problem of retrieving relevant images. In image mining, the goal is the discovery of image patterns that are significant in a given collection of images. This is certainly not true because there are important differences between relational databases versus image databases.
A. Absolute versus relative values In relational databases, the data values are semantically meaningful. For example, age is 35 is well understood. However, significant unless the context supports them. For example, a grey scale value of 46 could appear darker than a grey scale value of 87 if the surrounding context pixels values are all very bright.
B. Unique versus multiple interpretations A third important difference deals with image characteristics of having multiple interpretations for the same visual patterns. The traditional data mining algorithm of associating a pattern to a class (interpretation) will not work well here. A new class of discovery algorithms is needed to cater to the special needs in mining useful patterns from images.
DATA REDUCTION
As given, the input data of the proposed model is formatted as a set of equal sized raw and label image pairs. The transformation of the input image dataset into a database-like table and subsuming of the related features is described in this subsection. For the sake of clarity, various terms used for this process.
IMAGE MINING FRAMEWORK
Because of the image characteristics, pixels from a neighboring area will generate similar feature vectors in the transformation process. Under some circumstances, it will cause remarkable redundant information in the result table; for example, an image with a large portion of background.
Here we present some basic types of redundancy and show how they can be eliminated while converting the input image set.
1. Function-Driven Frameworks
These descriptions are exclusively application-oriented and the framework was organized according to the module functionality. For example, propose an intelligent satellite mining system that comprises two modules,
(a) A data acquisition: Pre-processing and archiving system which is responsible for the extraction of image information, storage of raw images, and retrieval of image.
(b)An image mining system: Which enables the users to explore image meaning and detect relevant events.
2. Information-Driven Frameworks
While the function-driven framework serves the purpose of organizing and clarifying the different roles and tasks to be performed in image mining, it fails to emphasize the different levels of information representation necessary for image data collection before meaningful mining can take place proposes an information driven framework that aims to highlight the role of information at various levels of representation. The framework, as shown in
(a) Pixel Level, also the lowest level - Consists of the raw image information such as image pixels and the primitive image features such as color, texture, and shape;
(b) Object Level- deals with object or region information based on the primitive features in the Pixel Level.
(c) Semantic Concept Level - takes into consideration domain knowledge to generate high-level semantic concepts from the identified objects and regions
(d) Pattern and Knowledge Level- incorporates domain related alphanumeric data and the semantic concepts obtained from the image data to discover underlying domain patterns and knowledge. The techniques frequently used include object recognition, image indexing and retrieval, image classification and clustering, association rules mining, and neural network.
IMAGE MINING TECHNIQUES
1. Object Recognition
Object recognition has been an active research focus in field of image processing. Using object models that are known a priori, an object recognition system finds objects in the real world from an image. This is one of the major tasks in the domain of image mining. Automatic machine learning and meaningful information extraction can only be realized when some objects have been identified and recognized by the machine. The object recognition problem can be referred to as a supervised labelling problem based on models of known objects. Specifically, given a target image containing one or more interesting objects and a set of labels corresponding to a set of models known to the system, what object recognition does is to assign correct labels to regions, or a set of regions, in the image. Models of known objects are usually provided by human input a priori.
2. Image Retrieval
Image mining requires that images be retrieved according to some requirement specifications. The requirement specifications can be classified into three levels of increasing complexity:
3. Image Indexing
Image mining systems require a fast and efficient mechanism for the retrieval of image data. Conventional database systems such as relational databases facilitate indexing on primary or secondary key(s). Now, the retrieval of most image retrieval system is, by text, similarity-based retrieval. In this case, indexing has to be carried out in the similarity space. One promising approach is to first perform dimension reduction and then use appropriate multidimensional indexing techniques. Indexing techniques used range from standard methods such as signature file access method and inverted file access method, to multi-dimensional methods such as KD- B tree, R-tree, R* -tree and R+-tree.
4 .Image Classification
Image classification and image clustering are the supervised and unsupervised classification of images into groups respectively. In supervised classification, one is provided with a collection of labelled (pre-classified) images, and the problem is to label newly encountered, unlabeled images. Typically, the given labelled (training) images are used to do the machine learning of the class description which in turn are used to label a new image.
Comments
Post a Comment