Research On Saliency Region Detection Algorithm Based On Feature Learning

With the development of image processing,visual inspection has been widely used in industrial automation production.In industrial visual inspection,there is a wide demand for accurate positioning of detected objects,and the subsequent processing of industrial visual inspection is based on the ability to accurately segment the detection area.The current saliency detection algorithms are basically aimed at natural scene images,and are rarely targeted at the industrial visual field.Unlike natural scene images,most of the detected images in industrial visual scenes do not conform to a priori information such as central priors and boundary background priors,and the foreground and background regions have similar feature similarities,so this paper is meaningful for the saliency region detection study of industrial visual scenes.In this paper,based on the characteristics of industrial visual images,a SPBL-based saliency detection system is proposed,which compares with the current better detection algorithms in industrial visual scenes and natural scene image sets,and has significant precision advantages.The main research of this paper content include:● Superpixel segmentation algorithm optimization:In order to speed up the processing speed of subsequent algorithms,the superpixel segmentation algorithm is applied to cluster adjacent pixels with similar features into one region,and because of the characteristics of industrial visual images,MBD is introduced to SLIC algorithm for optimization to solve the effect of noise on SLIC distance calculation.● Foreground and background similar image feature extraction:color information is an important feature to distinguish foreground and background regions.This paper focuses on the characteristics of different color spaces,and extracts information in images from different color spaces.The background in the image is generally richer than the texture features of the foreground,so LM feature and CS-LBP feature are extracted.In this paper,a 69-dimensional feature vector is extracted.● Classification optimization of SPBL based on structural information fusion:In this paper,the combination algorithm of AdaBoost and SPL algorithm is applied to the field of saliency detection.Boosting tends to reflect the local mode of data,which is more sensitive to noise data,and SPL tends to search for patterns in data with greater robustness,and the combination of the two algorithms draws on the advantages of both algorithms.Since the classification result of SPBL is the significance value of each super pixel region,the saliency value of each pixel is calculated by the structure information fusion algorithm,and the error occurring in the SPBL classification result is corrected.