Intelligent Image Technologies And Core Image Recognition System

Intelligent image processing technologies have been getting attractive in many fields. In engineering investigation, core recognition is an important basic task in geological exploration since accurate lithological identification can provide a reliable basis for further exploration and exploitation. Traditional methods are subjective and inefficient with low automation due to artificial gauge and laboratory analysis. Live real-time scanning of cores to acquire high fidelity images of the physical appearances for precise auto recognition of rock properties is of great significance in improving investigation efficiency, increasing analysis accuracy and intelligent operation level.The present content-based intelligent image recognition technologies face many challenges both in practical accuracy and concrete application. Based on the correlative domestic and international research findings as well as the characteristics and requirements in engineering exploration of core images, this paper studies appropriate image analysis and recognition algorithms to design and implement a core image automatic recognition system.In this paper, the focus is mainly put on the representation of image content and the classification of images. Meanwhile, images are grayed and median filtered to remove noises for preprocessing, considering the characteristics of core images.In terms of the representation of image content, the extraction methods of texture features and shape features of images are discussed. The extraction of texture features is done by statistical analysis method and Gabor filter method. The statistical analysis method uses the co-occurrence matrix to represent the texture features, which gives the gray-scale spatial dependence in image textures from a mathematical prospective. Gabor filter method uses a bunch of filters to filter the images, from which texture features is extracted. The extraction of shape features is done by shape description methods based on contour. Canny operator is used here to detect image edges.In terms of image classification, we focus on artificial neural network based BP (Back Propagation) algorithm and statistical learning theory based SVM (Support Vector Machine) algorithm as well as Hough transformation, which is used to recognize and measure the geometric shapes in images.Based on the studies of above mentioned algorithms, a core image automatic recognition system is designed and implemented, which is functional in image acquisition, image preprocessing, feature extraction and image classification.Testing of core images has shown that the recognition rate is more than 80%.