The Trademark Image Retrieval Based On Perceptual Hash Algorithm

With the popularity of the internet and information technology, the image hasbecome an integral part of daily life, it can present the information directly andvisually, and build a visual thinking mode which helps us to learn and think. How toretrieve the interested images from the mass of information quickly and accurately hasbecame a hot issue.Trademarks, which are the carriers of the products, are an important weapon forenterprises to participate in market competition. The competition between producersand traders is the competition of their goods, services or prestige, and is reflected inthe form of their brand popularity. The higher the brand popularity is, the morecompetitive the goods or services will be. All the factors will determine the trademarkhave to be protected by the laws, in other words, it must be compared with theregistered brands before a new brand is registered successfully, then decide whetherthe trademark is effective. However, it is not an easy task that retrieving the similarbrands from the mass trademark image database. Facing with such mass data, theearly traditional text-based retrieval method is not applicable obviously.The retrieval technique based on perceptual hash is a hot topic currently, thesearched results based on this technology are more coincident with human visualperception, it is able to generate a hash sequence which can mark the visualcharacteristics of the digital image through image feature extraction and quantizedcompression technology, then we can estimate the similarity via comparing with theimages’ hash. Therefore, the process of image feature extraction is directly related tothe efficiency and accuracy of the algorithm. We propose two improved methods inthis paper through analyzing the current problems of image retrieval techniques.The main work of this paper consists of the following aspects: (1)Research the current problems of the perceptual hashing technology deeply,then we design a perceptual hash algorithm based on image regularization by utilizingthe characteristics of image regularization. And through experiments, we demonstratethat the algorithm is with good robustness for rotation attacks, shearing attacks, JPEGcompression attacks and noise attacks.(2) Propose a method of getting perceptual image hash by utilizing blockstructure characteristics, and design a perceptual image hash algorithm based on areference image model. Through experiments, it is verified that the algorithm hasgood qualities of tamper detection, robustness and uniqueness.(3) Implement a practical application of perceptual hash technology——atrademark image retrieval system based on Java language, so that we can validate theperformance of the proposed algorithms in a more intuitive way.Finally, summarize the main research work and look into the focus of futureresearch.