Graph-based NAM Representation And Salient Region Detection

Image representation is a method of representation and storage in image information,which has important signification in computer vision, image processing and patternrecognition. The efficiency of image processing will be impacted by different imagerepresentation method. With the concept of packing problem, a non-symmetry andanti-packing pattern representation model(NAM) brings a well representation efficient,which can contain both a perfect image coding ability and a ideal image processing ability.In this paper, we focus on the need of image processing, propose a Graph-basedNon-symmetry and Anti-packing pattern representation Model(GNAM) based on NAM. InGNAM the patterns are split by global characteristic of graph and a local dynamic threshold.We design a method to calculate the distance between different sub-patterns for color imageand grayscale image. With the experimental data reconstruction by GNAM, the sub-patternsmade by GNAM can be more precise and the reconstruction effect can be better.Against the characteristic of the storage model of NAM sub-patterns queue and thecharacteristic of GNAM sub-patterns, we design a storage model called tag-matrix, whichcan contain the space characteristic and the compression characteristic of sub-patterns. Weimplement an algorithm of convert between sub-patterns queue and tag-matrix. Focus onthe characteristic of tag-matrix, we design a code compression algorithm combined byHuffman and run-length coding method. Analyzed to the storage model of NAM, wepropose a NAM coding protocol to form a unified regulation. Contrast to the data size ofthe sub-patterns queue storage model, the NAM storage model by tag-matrix is efficiency.This paper concentrates on the global and region characteristic of sub-patterns split byGNAM, an algorithm of salient region detection based on GNAM is introduced and computethe GNAM saliency map. Get the saliency cut by Grabcut algorithm used saliency map. Theanalysis and experiment shows that the salient region detection can gain a low time complexity.To contrast to other saliency detection method, our method has a higher precision and recall.The GNAM can efficiently used in image representation and image processing. It takes theoretical value and realistic meaning in image representation, image processingand image recognition.