Food Recognition System
Food image recognition has gained much popularity with the excessive use of electronic media where countless apps and devices concentrate on fitness and appearance and health.The main contribution of the proposed work in real-world is to determine any food item from a given image.
Once a food item is identified several apps in smart phones exist which can then list the number of calories it contains, the percentage of major food groups (e.
or any similar topic only for you
g.: carbohydrates, fats, proteins, dietry fibre, minerals, vitamins etc). Calorie counting and keeping track of the food one eats is of great significance to patients of diabetes, blood pressure, liver and gall bladder problems and also for athletes who need to keep track of specially their protein and carbohydrates intake.
A lot of people suffer from low body weight or are obese; both of them can mitigate this issue by keeping track of the daily calories. If such a food recognition system is employed in conjunction with the calories tracking apps it would be a lot of help to anyone who is conscious of their health and want to make better food choices.
This is particularly helpful because it’s quite a tedious task to input food item manually and if it’s a dish with many ingredients it gets even more time consuming to track the calories and the percentages of the major food groups in a given food item as compared to when one can automatically recognize the food item and retrieve it from the dataset provided.
When food is categorized according to nutrition, it has five main groups i.e.
- Milk and dairy products.
- Fruit and vegetables.
- Fats and sugars.
While according to the food guide pyramid introduced by the United States Department of Agriculture in 1992 food is classified into six major groups: protein, dairy, grains, oils, fruits and vegetables. However, in this paper we have used the FOOD11 dataset in which food is classified into 11 categories with each category containing different number of images. The food is categorized as follows:
- Bread(includes Bread, burger, pizza, pancakes)
- Dairy product( includes Milk, yogurt, cheese, butter)
- Dessert(includes ice cream, cookies, chocolates)
- Egg(includes Boiled and fried eggs, and omelette)
- Fried food(includes French fries, spring rolls, fried calamari)
- Meat(includes Raw/cooked beef, pork, chicken, duck)
- Noodles/Pasta(includes Flour/rice noodle, ramen, and spaghetti pasta)
- Rice(includes Boiled and fried rice)
- Seafood(includes raw/cooked Fish, shell fish, and shrimp)
- Soup(includes different kinds of soup)
- Vegetable/Fruit(includes Fresh/cooked vegetables, fruits and salads)
The FOOD11 dataset is collected from other existing food datasets like UEC-FOOD-100, UEC-FOOD-256, FOOD-101 and images from social media like instagram and twitter. This dataset is divided into three subsets of training, evaluation and validation. The total number of images in food11 dataset is 16643.
CATEGORY TRAINING VALIDATION EVALUATION
Bread 994 362 368
Dairy Products 429 144 148
Desserts 1500 500 500
Egg 986 327 335
Fried Food 848 326 287
Meat 1325 449 432
Noodles/Pasta 440 147 147
Rice 280 96 96
Seafood 855 347 303
Soup 1500 500 500
Vegetable/Fruit 709 232 231