Study On Source Cell-phone Identiifcation Based On Pattern Noise Of Sensor

In this era of rapid development in the information technology, digitalphotograph becomes an important carrier for information transmission, which notonly let us understand the trend of the latest information, but also becomes the validevidence in judicial and other areas.However, in the case that image tamper isbecoming serious, people doubt the authenticity of some sharing photographs, andstart to focus origin of the image. As a branch of image forensics, the sourceidentification is to explore the origin of the imaging device for a certain image, andbecomes a hot research direction.In the paper, effective methods are employed to identify photos captured bydifferent cell-phones.Began with a typical camera imaging model, we analyse thecharacteristic of each module from which forensics features are extracted. Based onanalyzing the existing several algorithms of image source forensics, two newalgorithms are proposed, and the specific research contributions are listed as follows:(1) An improved classification algorithm based on image features is proposed. Asthe color features of traditional image features have strong dependence on the imagecontent, so these features are removed in this paper, moreover, the image qualityfeatures and wavelet domain features are improved.In order to reduce thecomputational complexity, SFFS algorithm are employed for feature selection and itcan remove the redundant features which have little contribution to classification.Analysis and experimental results show that the new algorithm improves the defectsof the typical algorithm and classification accuracy is raised.(2) Correlation detecting algrithm of using three color channels is put forward.The algorithm firstly decomposes the images into red, blue, green three color channels,and three detection noises are acquired by denosing and de-CFA interpolation noiseoperation successively to each color channel.Then the method of canonical correlationanalysis is employed to identify images. Compared to traditional correlation detecting algrithm of using a single channel, this algorithm makes full use of the characteristicsthat sensor is the main source of noise and canonical correlation analysis canmaximize the correlation, so that the probability of the original photo matching realcamera is increased. The experimental results and analysis show that the algorithmhas good identification performance.