Obesity has been a severe public health challenge to the general population and social welfare in many developed countries. In the past three decades, the obesity rate in has increased significantly, resulting in serious consequences such as diabetes, stroke, heart disease, and even cancer. Food intake assessment is significant for obesity management. However, few people are aware of their food intake, and most are not willing to assess food intake. The reason is the burdensome assessment methods and a lack of real-time feedback with these methods. Traditional food recording or food diary methods require manual records of the food type and the portion of the food taken, and the accuracy is influenced by human estimations of the food portions. Computer-aided and automatic food intake assessment methods provide a convenient channel for food intake monitoring, although robust and reliable methods are not available for real-time field applications.
Food recognition is a special case of category recognition for visual
object recognition in computer vision. The appearance of any particular meal is affected by many factors such as ingredients, cooking methods, cutting patterns, ingredient positions, occlusions, lighting conditions, etc. These factors are complex such that even meals of the same category may have different appearances. In contradiction, different types of food could have similar appearances that are difficult to distinguish by humans. These intraclass differences and interclass similarities make food recognition a challenging category recognition problem.
|