The Region Of Interest Detection Algorithm With Multi-Feature And The Application In Different Scenes

The Region of Interest(ROI)detection algorithm of image has become a research hotspot.The detection methods of different scenes are different.A new ROI detection algorithm is proposed and improved for QR code beautification and video frame watermarking block selection,the main innovations are as follows:(1)A new algorithm for ROI detection based on multi-featureMost of the ROI detection algorithms are based on the Human Visual System(HVS).The single feature of the image is used to detect the ROI,and the detection accuracy varies with different application scenarios.Therefore,on the basis of the analysis of image features,combines the brightness and color features,the texture features are added to achieve a new multi-feature combination.The texture features are described by entropy,energy,contrast,homogeneity and correlation.The samples are divided by the texture complexity of image.And the weights are predicted by linear regression.The new multi-feature combination makes the detection of ROI more accurate.A better detection result can be obtained for the image with a large change in brightness.(2)A QR code beautification strategy based on ROIThe aesthetic QR code is generated by replacing the background image with the standard QR code,and the decoding rate is not impacted.The more ROI of the image is displayed,the better the beautification effect is.Since the existing algorithms only modify the data area,the replacement area is smaller.So this paper uses the proposed algorithm to detect the image.According to the ROI of single-target and multiple-target,the Positive Basis Matrix and the Selection Basis Matrix are improved to modify the data area and parity area,making the ROI can replace a larger area.And then the aesthetic QR code with better visual effects is obtained.(3)Block selection of video frame watermark based on ROIIt is the key problem to achieve the balance between transparency and robustness in video frame watermark algorithm.Some blocks of video frame are selected to embed watermark,and the selection of block is important.The multi-feature combination based on transparency and the multi-feature combination based on robustness are constructed to detect the ROI of block.The block is selected as the watermark embedding region when the both values reach their corresponding thresholds.Based on transparency,the salient features and texture features are selected for the multi-feature combination.The weights are obtained by linear regression.The positive and negative samples are calibrated by Peak Signal to Noise Ratio(PSNR)before and after embedding watermark.The multi-feature combination based on robustness not only selects the above features,but also the blur features are added.The corresponding weights are calculated by the linear regression learning samples.The positive and negative samples are divided by the ability of video frame to resist compression attacks.