WO2018190778A1 - Inspection procedure of the authenticity of brand-named leather bag - Google Patents

Abstract

This invention comprised of an algorithm to inspect the authenticity of any brand-named bags from its images by cropping authenticated images of the brand's trademark and the seam into an appropriate size using receiver operating characteristic, then slide the images from vertically and horizontally until a right overlaying percentage. After that, all the images will be analyzed and evaluated its authenticity from a deep learning method processes.

Description

INSPECTION PROCEDURE OF THE AUTHENTICITY OF A BRAND-NAMED

LEATHER BAG

TECHNICAL FIELD

[0001] The invention is related to Computer Engineering, majoring in Artificial Intelligence and Image Processing.

BACKGROUND ART [0002] Nowadays, Investor have extensive investment options other than bank deposit, bond, mutual fund or stocks which are real estate, gold, and rare commodity, in particular; collectibles antiques, drawings and brand-named products. In which trading brand-named products have been gaining a great deal of interest since it is highly profitable than others form of investment. The example of such a product are Chanel, Christian Dior, Christian Louboutin and specifically the classic Hermes Birkin and Kelly.

[0003] According to a recent research by Baghunter, the market value of a brand-named goods increases by 500% in 35 years, which equivalate to annual return of 14.2%, outperform S&P500 gold and stock index which have average annual return of 1.9% and 1 1.7% respectively. Moreover, the market value of a Hermes Belkin has neither dropped or fluctuates like those of stock and gold since the offer exclusivity only for selected customer that can't be brought by money. As a result, branded-name bag collection and second-hand trading has become increasingly popular among new investor.

[0004] Since the growth of brand-named bag price has attracted much supply and demand, there are group of people wanting to gain profit from trading of forgery product which will heavily affect buyer from social media platform such as Facebook, Line and Instagram etc.

[0005] In addition, forgery and counterfeit is considered as infringement of Thailand's Copy Right Act which cover literature, drama, art, music, audiovisual, cinematographic, sound recording and trademark. Currently, there are emerging numbers of counterfeit good in the market since more consumer want to increase their social status by purchasing in unaffordable branded goods like watches, accessories, and bags.

[0006] The production and distribution of these fraudulent goods should be controlled and inspected strictly. Consequently, an efficient and accurate tool should be developed to aid government officer in verifying the authenticity of trademark.

[0007] Genuity of an object can be inspect by image processing because the material quality and manufacturing process are not visible and can't be easily differentiate even if the comparing object are identical. To the best of our knowledge, there is no research or patent in Thailand that propose several embodiments to verify the authenticity of a leather bag to reduce copyright infringement and protect consumer right. International institute has been developing a tool and method to verify object in the financial and medical field. However, there is no known application for validating an object using image processing.

[0008] The research of published paper and patent are as follows;

[0009] United State Patent, US2012/0308124, a trained detector to locate fiducial points corresponding to parts within the image using labeled image. An input image is analyzed using the trained local detectors, and a Bayesian objective function is derived for the input image from the non-parametric model and detector scores. The Bayesian objective function is optimized using a consensus of global models, and an output is generated with locations of the fiducial points labeled within the object in the image.

[0010] United State Patent, US2013/0284803, proposes an integration of imaging system, database, and authentication center to configured unique signature associated with a good at the good's origin from its image. The unique signature can be, for example, a random structure or pattern unique to the particular good. The imaging system is configured to process the image of the good to identify at least one metric that distinguishes the unique signature from unique signatures of other goods. The database is configured to receive information related to the good and its unique signature from the imaging system; and is configured to store the information therein. The authentication center is configured to analyze the field image with respect to the information stored in the database to determine whether the unique signature in the field image is a match to the captured image stored in the database. However, a significant difficulty of the image processing was later resolved by United States Patent, US2010/0183217, which proposed a technique of employing a convolutional network to identify objects in images that are noisy or have areas narrow in color or grayscale gradient in an automated and rapid manner. For example, embodiment trains a convolutional network including multiple layers of filters. The filters are trained by learning and are arranged in successive layers and produce images having at least a same resolution as an original image.

[0011] Counterfeit banknotes and other form of forgery that are apparently noticeable have become attentive for researcher to propose detection embodiment, system and tool. It can be summarized as follows;

[0012] United States Patent, US5533144, A counterfeit detector and method for identifying a platen image portion to be photocopied contains monetary note patterns. The detection is performed in a rotation and shift invariant manner. Specifically, the pattern can be of any orientation and at any location of the image. Moreover, it can be embedded in any complicated image background.

[0013] United States Patent, US2008/0069424, a method for characterizing texture of areas within an image corresponding to monetary banknotes includes dividing the image into a plurality of sections; calculating a gray level for each section; selecting potential sections from the sections, the potential sections having gray levels within a predetermined range; selecting bill sections from the potential sections, the bill sections having pixels within a predefined color range and a predefined continuous color gradient range.

[0014] United States Patent, US2004/0213448 Apparatus and method for recognizing counterfeit currency by capturing a digital image of a currency bill to be recognized and a data storage for storing parameters, features, weights, and a feature identification instruction, feature- capturing unit for capturing features of the bill, neural network recognition unit for comparing the features of the bill with that of an authentic bill by performing an back propagation algorithm and using a plastic perception network as a training kernel, and output means for displaying a comparison result.

[0015] United States Patent, US0288182/2012, a mobile communication device and method for identifying a counterfeit bill are provided. In the method, an InfraRed (IR) image of a bill is received, feature values are extracted from the IR image to represent edges, a binary image having displayed pixels and non-displayed pixels is generated based on the feature values, and a corrected image is generated by overlaying the binary image on a pre-stored real bill database and adjusting the binary image to match a predetermined area of the binary image matches to the predetermined area of the real bill database. A number of displayed pixels in the corrected image is counted, and whether the bill is counterfeit is based on the number of displayed pixels in the corrected image. Apart from monetary counterfeit, more research has been made on detecting fake pill which could contain hazardous material. The proposed method as follows;

[0016] United States Patent, US2012/0013734, a device and method for the detection of counterfeit pharmaceuticals by visual inspection upon exposing a suspected counterfeit pharmaceutical to one or more light sources having different wavelengths and observing the differences in color or brightness between the suspected counterfeit and a genuine pharmaceutical/packaging.

[0017] To the best of our knowledge and according to research and patent mentioned above, there is no tools or method of verify the authenticity of a brand-named bag using partial image processing that are distinct in term of the learning and developing method with high efficiency and compatible to smartphone and tablet.

SUMMARY OF INVENTION

[0001] This invention associated with the inspection of the authenticity of a brand-named bag by using a distinct characteristic extracted from the material qualities and the production procedure to develop a reliable method with sufficient accuracy, sensitivity and specificity. The algorithm will inspect an image containing a segment of the bag that present with the brand's trademark that was compressed either the leather or the seam and the leather texture to be evaluated its authenticity using deep learning method.

[0002] The sole purpose of this invention is to prevent forgery and encourage legitimated trading of brand-named bag.

DETAILED DESCRIPTION

[0001] Figure 1 exhibit the program process. First, each image was input into the process of trademark and stich inspection (102) to confirm that the image is valid. The image will be cropped by receiver operating characteristic: ROC) so that the image contains all necessary composition. Program will slide the image horizontally and vertically continuously until the image achieved the optimal overlaying percentage. These data will be used by the deep learning method (105) for analysis and classification (108) of the bag authenticity and finally output (107).

[0002] Figure 2 exhibits steps and methods of processing with deep learning (105) we crop the image (104) and imported them in to our system. Then, we set variables such as filter values (201) the image will be classified into 6 types (202) and Gaussian Noise (203). The classified image will be processed by the following steps; convolution layer (204) function activation (205), max pooling layer (206) dropout (207) and fully-connected layer (208)

[0003] Figure 3 exhibit the analysis and classification (106) by counting the number of each type of image that was calculated from the deep learning method (301) if the ratio of authenticated image exceeds the authenticated percentage, the output will be 'authenticated'. For example, if the authenticated percentage is 70% (303), ratio of 70% or more will be output as 'authenticate' (304) and ratio of less than 70% will be output as 'forged' (305)

[0004] Figure 4 exhibits the architecture of deep learning processing.

[0005] Figure 5 exhibits the manual of the equipment by capture image of trademark, seam and leather texture, which can identify its brand and authenticity.

BRIEF DESCRIPTION OF THE FIGURES

[0001] Figure 1 exhibits the program algorithm.

Figure 2 exhibits the process and evaluation using deep learning method.

Figure 3 exhibits analysis and categorizing process

Figure 4 exhibits the architecture of process of evaluation using deep learning method

Figure 5 exhibits equipment manual.

Claims

1. Inspection procedure of a brand-named leather bag authenticity which consist of these procedures:

(a) receiving each image into the brand trademark and the seam to be cropped for a desired feature and appropriate size using receiver operating characteristic (ROC), the images will be slide horizontally and vertically throughout the image to calculate the authenticated overlaying percentage, this data will then be used for analysis and classification for the feature of the authenticated bag and output,

(b) the cropped images will be inputted and processed using deep learning method by defining multiple variables such as filter etc., and classification of the 6 types and increasing

Gaussian Noise, subsequently, the images will undergo significant processes including convolution layer, activation function, maximize pooling layer, reducing dropout and fully- connected layer,

(c) analysis and classification will be made by counting the number of images in each category proposed from the deep learning method and turn it into an authenticity percentage, if the ratio counted exceeds the output ratio from the deep learning method process, the output result will be authenticated, otherwise the output result will be forged,

characterized by

the deep learning algorithm is the process to differentiate authenticated bag from forgery bags by using feature selection from filter set, each filter set contain a fitting feature and size that was obtain from images of compressed trademark on the leather, stitching or texture that may not visible.