MATLAB 计算土耳其货币的硬币总金额.zip

### **Calculator of total money of coins in Turkish currency.** > **1.Introduction** -> The main goal of the project is coin detection and calculating the total amount of money in the given input image. I followed some steps to detect the coins which are preprocessing of the image that includes edge detection, binarizing, morphological operations, and creating an algorithm that is based on the reference coin. Reference coin in this project will be 1TL. And after determining reference coin, we sort the areas of the coins which detected in the image. After all, we create an algorithm according to the ratio which is “the area of the reference coin/sorted areas of the coins” with scaling it by “sqrt()” operator. With respect to the scaled values, we create the ranges between (0,1) and, we put the coin values on that ranges with some adjustment. Furthermore, we indicate the results on GUI application.GUI application is created by APP Designer. > **2.Preprocessing and Analyzing Image** -> To detect the coins we need some processes on the image.These processes are important because we need background and foreground relative to detect coins. These processes includes grayscaling the image, edge detection, binarizing and morphological operations.  > **3.Plotting Image Results** -> The result image after image prepropcessing.  -> The result image that shows the circled coins, their vlaues and the total money of the input image.  > **4.GUI Application Interface** -> GUI Application Interface images are shown below. > > **4.1- GUI applicaiton Interface** > >  > > **4.2- GUI after “Add an Image” button** > >  > > **4.3- GUI after “Preprocessing” button** > >  > > **4.4- GUI after “Calculator” button** > >  > **5. Strong and weak sides of the methods** > > **Weak Sides** > > > 1-) Changes in shadow, brightness and contrast in the background can cause some errors in our output. This can happen when using a certain threshold value. For example, we set this threshold value to 0.2, but it may not be appropriate for a different image. To avoid such errors and to be able to apply operations more generally, we usually adjust for input images whose background and brightness do not change much. That is, when choosing a particular threshold value, we have generally performed our operations with images where the background and brightness are similar. To improve suggestions : We can use machine learning to solve this problem. We can build a model for image segmentation tasks. We can train a model from a diverse dataset to learn the optimal parameters for different types of images. Evolutionary neural networks are particularly effective for these tasks. > > > 2-) Since the difference in area between 0.05 and 0.1 coins is very small, in some pictures 0.05 TL can be perceived as 0.1 TL and 0.1 TL as 0.05 TL.Also, the images where 1 TL’s are not in it does not work.Because we adjusted the reference point according to the 1 TL’s area. > > > -> The distortion effect due to background and not taking 1TL in the input image is shown below. > > >  > > **Strong Sides** > > > -> We reached very good efficiency with respect to the environment of the images we stated. Also we had very good results in the situations where the coins are overpallapping on each other. Our code still figured out detecting the coins. > > >  > > > -> RMSE = sqrt [(Σ(Pi – Oi)²) / n] Here, Pi denotes the predicted value, Oi represents the observed value, and n is the total number of observations or data points. The sum of the squared differences between the predicted and observed values is divided by the number of observations, and the square root of the result is taken to yield the RMSE.According the 39 input image in our dataset we have obtained RMSE value as “0.29”. This is very good result for an experiment.