Computer Vision News - February 2017

Say you have a large set of images to be used for classification or that you want to pre-process efficiently? How this can be done efficiently in Matlab? The answer is “DataStore"! In Matlab, a Datastore is an object created to streamline reading a collection of files or data. The Datastore acts as a repository for data that share the same structure and formatting. Datastore is useful mainly when each file in the collection might be too large to fit in memory. Matlab includes several types of Datastore, each having properties which are relevant to the type of data it supports. For example: • TabularTextDatastore - Text files containing column-oriented data, including CSV files. • SpreadsheetDatastore - Spreadsheet files with a supported Excel® format such as .xlsx. • ImageDatastore - Image files, including formats that are supported by imread such as JPEG and PNG. This month, we will demonstrate two uses of imageDatastore in two scenarios: 1. Image Classification 2. MapReduce Image Classification: We will show in the following example how you can load a set of images (belonging to two different categories) using ImageDatastore and classify them using bagOfFeatures and the SVM multiclass classifier. The set of images are (found in Matlab's vision toolbox): The set of images include two categories: cups and books. The book images are in the folder books, and the cup images in the folder cups, as can be seen below: Matlab's Image DataStore Tool Computer Vision News Tool 17