## Reactiv'IP

4 stars based on 53 reviews

The Most well known way to detect objects are using Deep Learning as it generates a fantastic result. However, Deep Learning Models need tons of data to train. So, for simple objects, we can go the traditional Computer Vision way to detect objects.

Images are group of pixels. Each pixel have a intensity value. So, we can treat image as just a matrix. By manipulating this matrix we can achieve a lot of interesting things. Blur an image can easily be done. We can take a small rectangle AKA Kerneland use this rectangle to slides through the entire image, while it's sliding, convolutions between this rectangle and convex hull binary image edges original image are taken place.

The convolution in image processing is just an element wise multiplication followed by a reduce summation. By doing convolution, we actually mix blend the nearby pixels together, therefor, blur is applied. If you think about how to define edges, you may come up with a definition like this: Convex hull binary image edges edge is a line which separates very different colors.

In Computer Vision people are actually doing the same thing. They find gradients of the colors and take a threshold to figure out the edges. Contours are basically connected shapes on convex hull binary image edges binary image. Using Contours together with other methods we can achieve awesome stuffs. Find Convex Hull of a contour can help us to take out unneeded information.

For example, we only want to keep the corner information on detection tasks. Having all those algorithm in our pocket, we can now try to detect simple objects. First thing we want to do is to apply Gaussian blur on the original image to remove high frequency noise. Second step, we convex hull binary image edges Canny Edge on the blurred image to get the binary image we wanted. Convex hull binary image edges contours on that binary image, then apply Convex Hull on contours.

Last, we can find Approximate Polygons on these Convex Hulls, it that's the shape we want, we keep it and this is a detected shape. Your email address will not be published. Notify me of follow-up comments by email. Notify me of new posts by email. Skip to content Why not Deep Learning?