Project Summary IS

Calorie estimation from food images using Machine

learning

Calorie Tracker applications are popular for on-diet population. With such applications,
people can enter their meals and estimate the daily consumption of calories. Somehow, the
meal entry task can be inconvenient: sometimes it may not be possible to judge the size of
the dish (while eating out for example). Thus a good idea to help making these applications
more user-friendly is to ask users to take a photo of their meal, and from that photo to give
a calorie estimation.
The project consists of two steps: identifying food from an image, and convert the food
identification into a calorie estimation. They perform food image classification using SVM
and deep learning algorithms. Different features such as LBP, HOG, and CNN are explored
and compared. For the calorie estimation step, they create a “calorie map” for the image
classification labels. The dataset is already available with images of food items and their
classes. This dataset is processed i.e. cropped to an appropriate size and image processing
algorithms like LBP and HOG are performed on those images to capture the features about
the geometry and texture of the food item. These features are inputted to different machine
learning algorithms and their confusion matrix is constructed. The accuracy and precision
are measured and the results are compared. CNN(fine-tuned) model is found out to be the
most accurate with an accuracy of 96.84%. The dataset has 10 categories and hence the
project team plans to scale up the project by taking a larger dataset and the same training
methodology.