A Silhouette Recognition System Combining Feedback ICM Neural Network and FPF

Abstract

A silhouette recognition system combining feedback ICM neural network and FPF, which uses the pulse coupling characteristics of ICMNN to extract the complete outline of the target image; and the feedback mechanism adopted continuously enhances the original image, so as to suppress non-similar images. The target enhances the purpose of similar targets at the same time; with continuous iteration, FPF continuously searches for candidate targets similar to the target edge in the image processed by ICM, and when a reliable similar target is found, a larger correlation will be generated at the corresponding position peak, and because the addition of the feedback mechanism effectively suppresses other non-target regions, the correlation values ​​of other regions are reduced, thereby achieving reliable recognition of the target. The system of the invention can better recognize images of the same category in the figure, and also has better stability for images with certain rotation and scale changes. Compared with other silhouette recognition algorithms, the calculation amount is reduced, the workload is correspondingly reduced, and the speed is increased.

Description

technical field [0001] The invention relates to a silhouette recognition system combining a feedback ICM neural network and an FPF, and relates to the technical field of image recognition. Background technique [0002] Silhouette recognition is an image recognition technology based on object outlines, which is widely used in robotic grasping, medical image processing, and content-based image retrieval. Over the years, there have been many silhouette recognition algorithms. The classic algorithms are as follows: 1) Statistical learning-based methods. Statistical learning-based methods are generally divided into a learning phase (for training classifiers) and a classification phase. The learning stage is to evaluate the distribution of each object edge in the feature domain under different category modes; the classification stage is to use the object edge distribution obtained in the evaluation stage to classify the silhouette image. This type of method relies on a large numb...

AI Technical Summary

Problems solved by technology

If this type of method does not use the exhaustive method to calculate the connection weight in advance, it cannot use the method of step-by-step recursion from the back to the front that is commonly used in dynamic programming, so the amount of calculation is very large.

Images