Category
JFL, Active Learning Classroom (171)
Description
The functional performance of porous metals and alloys is dictated by the pore features such as size, connectivity, and morphology. Although methods such as mercury porosimetry or gas pycnometry can provide cumulative information, direct observation using scanning electron microscopy (SEM) often provides important details unavailable through other means, especially for microscale or nanoscale pores. Hundreds or even thousands of pores may be present in each scanned image, and efficient identification, classification and quantification of pore characteristics require automated processing of the pixels for edge recognition. In current practice, the pores on scanned images were hand-traced for outlines and imported into software to isolate and analyze the pores using thresholding, contrasting, and binary morphological operations. This approach was unable to consistently capture both large and small pores while accurately removing noise. In this paper, a software framework was designed and implemented, taking advantage of modern computing tools, libraries, and methodologies for automated image processing on pore identification, classification, and quantification. Vectorization was adopted as the key technique for edge detection using both the magnitude and directionality of the detected edges. This technique allowed for broken or incomplete edges to be recovered into complete pores, while filtering out the noises. This automated image processing tool greatly reduced the manual work and improved the speed of pore analysis while maintaining high-level of accuracy in pore metrics.
Porus: An Automated Image Processing Tool for Scanning Electron Microscopy
JFL, Active Learning Classroom (171)
The functional performance of porous metals and alloys is dictated by the pore features such as size, connectivity, and morphology. Although methods such as mercury porosimetry or gas pycnometry can provide cumulative information, direct observation using scanning electron microscopy (SEM) often provides important details unavailable through other means, especially for microscale or nanoscale pores. Hundreds or even thousands of pores may be present in each scanned image, and efficient identification, classification and quantification of pore characteristics require automated processing of the pixels for edge recognition. In current practice, the pores on scanned images were hand-traced for outlines and imported into software to isolate and analyze the pores using thresholding, contrasting, and binary morphological operations. This approach was unable to consistently capture both large and small pores while accurately removing noise. In this paper, a software framework was designed and implemented, taking advantage of modern computing tools, libraries, and methodologies for automated image processing on pore identification, classification, and quantification. Vectorization was adopted as the key technique for edge detection using both the magnitude and directionality of the detected edges. This technique allowed for broken or incomplete edges to be recovered into complete pores, while filtering out the noises. This automated image processing tool greatly reduced the manual work and improved the speed of pore analysis while maintaining high-level of accuracy in pore metrics.
Comments
Graduate