A preliminary study on the microscope imageTo complete a machine vision project, it is necessary for the developer to understand the image, analyze the features, and put forward the initial ideas. For three microscope images, the original requirement was to be able to identify three types of objects with a 75% success rate. Look at the image first
first, the subjective experience of the image1) from the overall color, found that three images of the main color is not uniform (Figure 1 Purple, Figure 3 green), should be caused by the different materials of the slice. If all the slices belong to this hue, you can first consider strengthening the ROI region through the color gamut. This requires communication and more material;2) from the image content, can be divided into 1. Background, 2. Large area, 3. Small bubbles, 4. Big bubbles. 4 regions. Whether the division is correct, what each region represents (which requires expertise), requires further communication. 3) from the image features, can be divided into 1. The background area is a large area of continuous area, there is fuzzy interference; 2. Large area features are obvious, black pixels occupy a large space, and 3 and 4来 say that the characteristics of the bubbles are as follows: and basically a circular or elliptical state.
second, the initial analysis of the image1) Color and area analysis, simple can be seen in the color blue region and V region, background and image has a good distinction2) Histogram analysis, three of the original image of the histograms are basically occupy all the space, while the foreground background is not obvious, can not be removed by a simple background removal Method 3) Frequency domain analysis of the FFT image, no obvious texture features, the image directionality is not obvious.
three, simple image processing1) Edge Analysisafter the Sobel operation, it can be found that most of the edge points have been looked for. This is especially good for bubble areas.2) cluster analysis for large areas of the same area, can be directly determined by the method of the block analysis of the region. Then the related morphological operation, can get the final result above all just to complete an image recognition of the initial processing. Successful machine vision projects rely on accurate requirements understanding, efficient communication, problem solving algorithms and the right hardware platform. But through the preliminary treatment here, basically can determine the work after.
A preliminary study on the microscope image
