| For a long time,the fake products have done serious harm to the world economy and consumers.Therefore,the anti-counterfeiting labels of the products have been paid great attention by people,among which the anti-counterfeiting mark is the most common means.The usual anti-counterfeiting marks refers to the labels with anticounterfeiting effect that can be pasted,printed,and transferred on the products’ surface or packaging.Traditional methods include direct vision,touching,texture,perspective and instrument detection However,these methods are either highly unreliable or inconvenient to operate.With the rapid development of smart phones,the acquisition of digital images is becoming easier and easier.As a result,to identify the authenticity of the anti-counterfeiting marks by using digital image technology has become a fast and convenient means with a certain degree of reliability.Therefore,it is urgent requirement for the sake of social economic development to work on the automatic identification of the anti-counterfeiting pattern authenticity and its system based on digital image.The relevant technology and existing difficulties of the anti-counterfeiting patterns identification based on digital images have been summarized in this paper,and studies a kind of anti-counterfeiting pattern composed of randomly generated texture pattern,to ensure that each anti-counterfeiting pattern is different,and its unique subtle grain and original printing physical features can be very useful against forgery.Although the forged anti-counterfeiting pattern will have differences in stroke weight,shape and position,the same true anti-counterfeiting pattern image collected by different devices and environments will also have slight differences,making it difficult to accurately distinguish the true and false images based on the differences directly for the carefully forged anti-counterfeiting pattern.However,the forged anti-counterfeiting pattern inevitably has a forged trace making it different from the true anti-counterfeiting pattern in the subtle region of the anti-counterfeiting pattern,which makes it possible to identify the anti-counterfeiting pattern.Therefore,a set of automatic recognition algorithm has been proposed,including anti-counterfeiting image registration,anticounterfeiting image segmentation,anti-counterfeiting image micro-features and texture feature extraction and authenticity identification.The main work and innovations are as follows:A.Aimed at the registration problem between the anti-counterfeit images to be tested and the sample images,a method based on feedback grid clustering is proposed to register the to-be-tested seal and the reserved seal.First,the ORB features of the anticounterfeit images to be tested and the sample image are extracted and initially matched by using a brute force matching algorithm.Then,the images are divided by a unilateral grid,and then the local feature cluster is used to perform coarse filtering of the fast feature points,which eliminates a large number of mismatches enabling us to finish numerous mismatched filters within milliseconds.In order to ensure the correct matches not mistakenly filtered out,we further design a local linear transformation to conduct feedback verification.A precise screening has been done on the correct matching points deleted accidentally and the false matching points not deleted in and around this region.The registration method utilizes the spatial consistency of the feature points registration to most effectively eliminate the mismatch,not only extracting high correct rate matching from the coarse baseline matching with low correct rate,but also retaining the correct matching points to the greatest extent,which realizes the accurate registration of the anti-counterfeiting images and sample image.B.Aimed at the accurate segmentation problem of the anti-counterfeiting images to be tested,a local threshold segmentation method based on sample supervision and guidance is proposed.The difference region between the anti-forgery binary images to be tested is extracted according to the binarization effect in the sample images.Then the weighted local mean deviation of the pixels in each of the different regions are calculated for adaptively selecting the optimal segmentation threshold,which is applicable to each of the different regions,and also verify the binary results of the difference region.The experiments show that this method reduces the quality requirements of the anti-counterfeiting images to be tested,and can restore the characteristics of these images to the greatest extent,effectively decreasing the differences between the genuine anti-counterfeiting binary images and the sample binary images due to the distortion of pen bond and incompletion.C.Aimed at the detail identifiability problem of the anti-counterfeiting images to be tested,a method for automatically identifying the authenticity of the anticounterfeiting images is proposed based on the quantitative difference of the key nuances.The difference between the intentionally forged anti-counterfeiting pattern and the sample anti-counterfeiting is small,while the true anti-counterfeiting pattern varies slightly in different devices and environments.In order to accurately distinguish the two differences,it is proposed to extract the skeleton shape in the sample binary images and calculate the bone width by using the bone width transformation,and then extract the narrow bone as the key region,and the anti-counterfeiting images to be tested and the sample images are compared according to this key region,and the difference values are calculated to identify the authenticity of the anti-counterfeiting image to be tested.The Experiments point out that the accuracy rate is higher than 97%and the recall rate is also over 90%.Also,the accuracy rate can be increased to 100% based on the actual application requirements with the recall rate only decreasing slightly.D.Aimed at the texture identifiability problem of the anti-counterfeiting image to be tested,a new texture feature extraction method,namely the Circumferential Local Ternary Pattern(CLTP)is proposed,significantly improving the recognition effect compared with the traditional method.Firstly,an anti-counterfeiting key point region composed of a chamfer or a small gap region is constructed according to the characteristics of ink jet printing.This paper adopts SUSAN feature points combined with subtle key areas to construct the anti-counterfeiting key points region so as to quickly locate this anti-counterfeiting regions.Then,the CLTP texture features of the anti-counterfeiting images to be tested and the sample images in these regions are calculated to form a feature histogram vector.The authenticity recognition is performed according to the similarity of the feature histograms of the anti-counterfeit images to be tested and the sample images.The experimental results show that the identification accuracy rate can be as much as 100% by using the anti-counterfeiting key points region combined with CLTP characteristics.It is also experimentally proved that the method has high discrimination,stability and validity for the authenticity identification of digital anti-counterfeiting images.With the in-depth study and exploration on the authenticity of anti-counterfeiting patterns based on digital image,this paper makes full use of the characteristics difference between the forged anti-counterfeiting patterns and true anti-counterfeiting patterns from various aspects,proposing an effective identification method of the digital anti-counterfeiting pattern images with comparatively better recognition effect.The research work of this paper makes the authenticity recognition technology of the digital anti-counterfeiting images more practicable,which is of great significance for protecting the products intellectual property rights and maintaining the stability of the sociality. |
