Examples of parametric models are: lines / circles/ arbitrary shapes/templates. As an easy example: Line. Line model has three requirements: 1. how many lines are there? 2. which points combine which line? 3. what is the line? Parametric models use voting systems to fill in occlusions and cancel noisy lines,Continue Reading
Smooth the image with a gaussian filter to remove noise. Find gradients with max intensity (above a certain threshold). Apply non-maximum suppression to get rid of spurious response to edge detection (by finding the peaks in the gradients detected above the certain threshold, so it basically is the maximum ofContinue Reading
First when you just have 1D and want to find edges such as just a vertical or horizontal scanning, it’s easy to take the second derivative times the kernel and times the image: ∂²/∂x² * h(kernel) *f(image) for horizontal or ∂²/∂y² * h(kernel) * f(image) for vertical. Then just find aContinue Reading
An image gradient is a(or the most) directional change in the intensity or color in an image. * Canny edge detector uses image gradient for edge detection. The gradient of a two-variable function (here the image intensity function) at each image point is a 2D vector with the components givenContinue Reading
Regression is estimation(correlation) – of the independent variable with the dependent variable.(We can make a scatter plot whenever you have data that relates two values together). We need an equation – which models the data for y(dependent) for any x(independent), that is, the correlation between them. If the data pointsContinue Reading
Given a line ‘a’ and a point ‘b’ we want to find a scaled version(λ)’a’that is as close to ‘b’ as possible. The right answer is where the line of λa to b is at a right angle to line a. This line is the difference of the line λa and theContinue Reading
Bayes’ Theorem – Is the probability Event A will happen given (|) Event B happened. P(A|B) = P(B|A) * P(A) / P(B) For example (Assume we have 3 out of 4 probabilities): Calculate a probability that an event A(Eating) will happen given (|) that an event B(Exercising) happened. P(A|B) = P(B|A)Continue Reading
Computer machine rounding errors produce values that are close to zero, but are not zero, thus they increase the ranks of a matrix when they shouldn’t. In order to solve this problem a threshold is determined so that the values become zero if they are below the threshold and thusContinue Reading
Decrease the learning rate. Increasing (multiplying) the number of filters in each convolution layer in the ANN. Add extra convolutional layers (can duplicate the existing ones). Add more DropOut layers. All , and any of these steps lead to improving generality accuracy of the ANN model.Continue Reading
Get the supervised dataset (containing the image features and their labels). Validate the data – by size , by shape, by error data , by premature EOF, by Null type, by object type, by out of range of enumerated values, by business logic. Visualize the data by plotting it. CreateContinue Reading