Based Watershed Segmentation Adhesions Particle Image Analysis Technology

Particle was common analyzed and measured object in industry, agriculture, medicine, scientific research etc. We could analyze their quality and characteristic by counting the numbers and extracting their features. Although manual observation could identify and classify different particles, it was complicated, inefficient. Adopting image processing to analyze and measure particles, which can easy work intension, enhance the efficiency and accuracy of work. As a result, it became a research hotspot in recent years.The main researches of this thesis include image pre-process and improve, image segmentation and counting, parameter analyses. Firstly this thesis introduces some pre-processing, such as converting image to grayscale image, it removed unwanted color information; Using median filtering algorithm for image smoothing to remove kinds of noise; Using Otsu algorithm to select the appropriate threshold for image binarization, this separated particles out from the background; Besides, this thesis designed a mathematical morphology filter, it filter out the irregular edges of the image, making reduced connection of particles and filled holes. Then propose an improved watershed segmentation algorithm. It used distance transform to generate a new grayscale image, eliminated a lot of "false" local minima in original grayscale image, then segmented grayscale image to make the touching particles well separated. Usually over-segmentation produced some areas with small area, we can select a area threshold to remove the small areas, and the division would become better. Large numbers of experimental results show that the algorithm can be applied to different areas of adhesion particle images, with good adaptability and robustness, faster processing speed. Finally, two kinds of particles automatic counting method were proposed, and particle size, shape, color and other parameters were used to describe the characteristics of particles, the characteristic parameters were extracted.Image segmentation and analysis algorithm in this thesis can separated touching particle accurately and quickly, the accuracy of particle counting reached 99% or more, the results of feature extraction was quite true. The algorithm processing is quickly, high accurately, good adaptability and real-time, it laid a good foundation for identification and classification of particles.