Research And Application Of Image Detection Method Based On Local Image Feature

Objective: Image detection is one of the important support technologies in the field of pattern recognition. It aims to utilize information technology to recognize images automatically. The mainly technology basis of the image recognition is image feature. With the development of local image feature technology and the effectiveness of image feature extraction and expression, the current image detection method based on local image features has been a hotspot of this research field.Based on the local image feature technology, this paper uses it into the interstitial lung CT image detection and human motion gesture image detection. The purpose is for the accurate detection on lung parenchyma of interstitial lung CT images and improving the accuracy and efficiency of human motion gesture tracking estimation and recognition.Methods: Based on the local image feature of lung CT image, this paper presents an interstitial lung CT image detection method. First, the method uses Turbopixels technology to do over segmentation to get superpixels image. Second, the gray co-occurrence matrix is used to extract the texture information of, combing the gray average and variance of superpixels in order to get the characteristic matrix. At last, random forest algorithm is used to get the detection segmentation result of interstitial lung CT images.Based on the local image feature of depth image, this paper designs a human motion gesture image detection method. First, the method use Kinect to obtain depth image and then converted the depth image into the skeleton point data. Second, the three step search algorithm is used to calculate the block matching of human skeleton image in order to estimate human motion gesture, at the same time, the Euclidean distance feature of skeleton point image is used to recognize human action.Conclusion: The research results show that the interstitial lung CT image detection method can overcome the influence of different performance types by interstitial lung disease on the accuracy of segmentation and the smoothness of the boundary effectively, having a good image detection segmentation result. The method has the segmentation accuracy reached on 98.07% in normal lung and 96.23% in diseased lung. The algorithm has better performance and certain application value.The experiments results show that the human motion gesture images detection method can overcome the influence of environmental factors such as illumination and express human motion gesture effectively. It has better accuracy and robustness in the human motion gesture estimation and action recognition. Meanwhile this method can simplify a series of calibration work of 3D video surveillance for observing the gesture estimation and action recognition of object in a real-time. This method has certain application value and reference value for similar work.