Image Classifications with Neural Networks

Digital images contain many contextual objects in them. Similar objects of different images show statistical data similar to each other despite the setting of any given image. We propose the use of Artificial Neural Networks (ANN) to classify images based on the object in the image. To achieve this goal, we will use ANNs to calculate probabilistic data in order to predict the classification of other like images. There will be pre-existing sample data which will be labeled accordingly. We will employ gradient descent to determine the best fitting boundary parameters. Gradient descent will find the minimum error boundary parameter. In order to ensure the accuracy of our classification, we will implement regularization. Regularization will prevent overfitting of the boundary parameter which in turn will allow for accurate boundary parameters. A technique called gradient checking will also be used to verify the calculations of gradient descent. To further develop our classification techniques, we will build an ANN capable of distinguishing objects without the use of pre-existing sample images. This type of ANN is called an autoencoder. It utilizes a single and multi-layered network to learn the process of reconstructing unlabeled data. After many inputs, the ANN will be able to label data on its own. This will achieve the ultimate goal of classification.