Food Ingredients Identification System Research Based On Theandroid

Recently, food safety has increasingly attracted people’s attention. There aremany ingredients on packages such as “Insoluble Polyvinylpyrrolidone” that peoplecannot understand. In the past, people might not dig deep about them. But withwidely-used internet, people can get information faster and more convenientlynowadays. They can even use a mobile phone to search for what they want. So farthere hasn’t been any special system for searching ingredients. What people do now isjust put an ingredient’s name in a search engine and then click “enter” to get theinformation. In this case, difficult words or words with many letters may bring a lotof trouble. Therefore, a system for obtaining information on ingredients convenientlyand quickly is of crucial importance.After analyzing the inquiry mode and character recognition system of currentingredients, this thesis proposes a food ingredient recognition system based onAndroid considering people’s habit of using smart phones. Users take photos of foodingredients with terminals with Android system and upload the photos to the server.The information about the ingredients will be shown on Android terminals after aseries of preprocessing and character recognizing procedures, which benefits people alot for obtaining ingredients information. In aspect of theory, this thesis discusses indetail the preprocessing, including linearization, tilt correction, charactersegmentation normalization, refining as well as feature extraction, categorizer design.As for practice, this thesis has finished the whole system’s design and application,which compared with existing inquiry mode and characters recognition system in theaspects of convenience, recognition effect and the comprehensiveness of results hasmade great achievements.The innovations of this thesis are as follows:1) On the basis of the controller and the features of ingredient sheets, this thesisuses projection methods and cluster analysis to get threshold value makingsegmentation technology separating words into single characters possible, whicheffectively extract characters and words of ingredients. This technology offers morekinds of modes and more accurate location than traditional single word extraction. 2) Due to the limitation of the database of ingredients, this thesis presents theintelligent error correction algorithm based on edit distance to improve the randomerror of characters recognition, which effectively solve the problem that is the rate ofcharacter recognition cannot reach100%.3) The strategy based on the recognition of whole words rather than traditionalsingle characters can provide more mode information and reduce dimensions, whichcan lead to more accurate recognition results and is instructional to other specificsubdivision fields.4) The thesis comes up with a strategy to learn from feedback according to thehabit of using mobile terminals. The more users employ the system, the moreaccurate their results will be.