Research On Image Recgnition Algorithms And Implementation On Intelligent Mobile Phones

Image recognition is to process image information and make classification of images automaticly by computers. Digital image processing and recognition is an important research field of Pattern Recognition. In recent decades of years, image recognition technologies have achieved deep and rapid development, and are applied in various fields such as image remote sensing, robots visual, biologic medicine and geological prospecting.Aiming for engineering applications of electronic recording system, in this dissertation, digital image recognition technologies are studied to be implemented on intelligent mobile phones. Several steps such as photographing, file saving, image pre-processing, features extraction, recognition and classification are implemented and the softeare are developed on mobile phones with the functionalitie s of automatic image input and recognition.The dissertation mainly contains three parts:research on image recognition algorithms, design and implementation of image recognition system on intelligent mobile phones, testing of ground environment images.Several common algorithms involved of in digital image recognition system are analyzed in detail in the dissertation. Considering time consuming and resource usage, proper algorithms are selected on mobile phones. Images are pre-processed with image graying and gray stretching. Image features such as color, texture and edge are extracted with gray values statistic, Gray Level Co-occurrence Matrix, Template Detector methods separately. Back Propagation Artificial Neural Networks and Support Vector Machine algorithms are implemented on mobile phones so as to establish the recognition and classification system.To improve the efficiency and practicality, some special considerations are discussed to fit for the limitations of mobile devices such as low CPU speed and small memory space. The operations include pixel information extraction via memory copy method, format conversion to JPEG of classified images, and transplant the sample training step to personal computers.The software is implemented in C# on Visual Studio 2005 platform based on.NET Compact Framework. Images of ground environment such as clay, grass and stones are used to test the system and the experimental results indicate that this system work well.