Food Image to the Recipe Generator

Food Image to the Recipe Generator

***

Abstract - We intend to use image classification to detect which food dish it is and recommend recipes based on that detection. When it comes to food, there are lots of options and unexplored dishes people might not be aware of. While there are applications that can recommend users with recipes that they search for, ours will use image classification to identify the food dish and suggest related recipes based on its image.

It is possible to capture the input image or to choose it from food images already present in the mobile device's camera. Deep learning convolutional neural network will be used for the classification, as well as for the recipe recommendation.

It

1.INTRODUCTION

People these days are increasingly aware of the importance of what they eat, so that they will make better nutritional decisionspertainingtotheirhealth,livelihood,emotions,andculture.Foodisafundamentalpartofthehumanexperience. It impactsourhealth,ourlivelihood,ouremotions,andourculture.Despitethis,manypeopledonotknowwheretostart.

Inordertohelppeopleachievetheirhealthgoals,wecreatedanapplicationtohelpthembetterunderstandandcontroltheir food consumption. We are creating a system that accepts an image of a meal as input, and then outputs recipes and nutritionaldatathatarecloselyrelatedtoit.Ausercanchoosefromseveralrelatedrecipesgivenaparticularinputimagein Snap N Eat,andeachoftheserecipesareaccompaniedbynutritionalinformationinordertoassistpeopleinmakingbetter nutritionaldecisions.

In addition to creating a more sophisticated understanding of health and diet, the platform could allow users to express themselveswithnewrecipesandcookingtechniques.

1.1 Dataset

DatasetsrelatedtoIndianfoodareincludedinthisdataset.ThemostcommonlyusedrecipeisselectedfromKaggle.

Thisdatabaseincludes4000imagescategorisedin80differentclassesorcategories.AsIndiancuisineconsistsofavariety of regionalandtraditionalcuisinesnativetotheIndiansubcontinent.Giventhediversityinsoil,climate,culture,ethnicgroups, andoccupations,thesecuisinesvarysubstantiallyanduselocallyavailablespices,herbs,vegetables,andfruits.

UsingJupyterNotebook,wecleanedandbuiltthemodel.

1.2 MobileNet Model

Intermsofmobilevision,MobileNetisaCNNarchitecturemodelforimageclassificationappliedtomobiledevices.Thereare other models as well, but what makes MobileNet different is that it requires very little processor power to run and applies transferlearning.

This makes it a great fit for mobile devices, embedded systems, and computers without fast graphics processors or low computational efficiency without sacrificing accuracy significantly. This method also works well with web browsers since theyarelimitedbycomputation,graphicprocessing,andstorage.

MobileNet

Model