Research On Variable Process Models In Application Of Recognizing Image Objects

The process of recognizing objects in images is a variable process with common processing links.If an image object recognition process model can accommodate these changes,it will not only reduce the number of process models and the workload of modeling,but also can improve the flexibility of image object recognition process.In this study,establish the process model of image recognition by using workflow technology and describe variability points in the process model based on constraint programming theory.The research contents include:(1)Research on variable points of image object recognition process.In the preprocessing link,the variable points include the selection of one or more kinds of image processing technologies(such as horizontal flipping,brightness enhancement processing,etc.),the selection of two categories of target detection models and semantic segmentation models,and the switching of two modes of training and verification of image object recognition.(2)Research on process model of image object recognition.The process model is divided into main process model and sub-process model.The main process model is a process model that abstracts the common links of image object recognition process processing into common activities.The sub-process model is used to express a variety of alternative processing technologies and corresponding processing processes for image object recognition.The main process model and the sub-process model can consist of variable activities expressing points of variation as well as non-variable activities.(3)Research on variable model of image object recognition.The variable model expresses the variable activities in a process model in a tree structure.The root node of the tree structure is the name of the process model,the node of the tree structure is the name of the variable activities in the process model,and the leaf node of the tree structure is the name of the sub-process model or the name of the function module.In the variable model,in order to reflect the requirements of constraint programming,nodes can set activation logic conditions for sub-process nodes,and sibling nodes can express dependency with constraints."An image object recognition variable flow model" is composed of "An image object recognition flow model" and "An image object recognition variable model",which can realize the integration of two modes of training and verification of image object recognition,and can add a training sample in a pre-processing link when the sample size of the training of the image object identification is insufficient.Aim at different characteristics of that image object,a target detection mode or a semantic segmentation mode can be selecte for recognition.In addition,based on the constraint programming theory,the integrality of the combination scheme of the model expressing the above functions can be evaluated.The research results provide a useful reference for the automatic implementation of image object recognition and the evaluation of function combination schemes.