Research On The Method Of Salient Region Detection In Fabric Image With Complex Background

Human visual system can quickly gain significant area from complex scene image and thereason is that visual selective attention mechanism works. By simulating the human visualselective attention mechanism to build visual saliency model, we can make the computer quicklyextract significant areas of complex scenes image in the absence of prior knowledge. At present,many visual saliency models have been proposed. These models predict significant area of naturalscene image from different angles. For fabric images with complex background, the effectivenessof these models is not ideal.Fabric image is different from natural scene image. The obvious characteristics of fabricimage are rich texture information, so the analysis for the fabric image is just for texture analysis. Ingeneral, Texture can effectively reflect the feature information of the fabric image, so people takethe textural information into account firstly in the detection of fabric image. The texturalcharacteristics of fabric image include the degree of thickness, uniformity, directionality andrandomness. Current algorithms have a high computationally complexity and are sensitive to noisein extracting features of these parameters. The main reason for these problems is that thebackground of the fabric texture image is complex. It is easy to treat defect free area as defect.Although textures is diverse and the type of defect is various for cloth, fabric defects understandhuman attention in the complex texture image. Fabric defect detection based on visual saliencymodel making using of human visual perception mechanism and the textural features of fabricimage is a promising research.Making using of textural structure feature, we propose a fabric defect inspection algorithmbased on the abnormal texture structure. The primary idea of the algorithm is based on theredundancy of pixels and we can get the dominant neighborhood structure map representing the global structure feature making use of the redundancy of pixels. We define the saliency of pixel bysubtracting the dominant neighborhood structure map by neighborhood structure map of the pixel.The results of our experiment prove that our method gets the desired effect and it is robust torotation, noise, change in scale, and change in luminance of fabric image. But when the image islarge, our method has a low efficiency of operation. For the above questions, we propose a fabricdefect detection algorithm based on local statistic features and global saliency analysis using LBPoperator to extract textural feature and global saliency model of selecting blocks randomly takingefficiency into account. The algorithm can detect defects fast in fabric image with complexbackground and requires no reference sample. So it has a strong adaptive capacity. Comparing withthe current saliency model, the proposed saliency model can efficiently distinguish the defect;moreover, segmentation scheme is superior to the current defect detection algorithm at detectionand localization.