Study Of Electronic Image Stabilization Algorithm Based On Feature Matching

With the progress of science and technology, all kinds of cameras can be seeneverywhere in the people’s life, such as mobile phone, camera, video camera, traveledrecorder and video-surveillance equipment, etc. But, because of a shaking of thecamera carrier, the video taken by cameras contain jitter, causes image blur, and flogof the video image can also make viewer feel visual fatigue. Therefore, video filmedwith irregular motion on camera carrier needs processing, removing unnecessary jitter,generating the image stabilization technology.Electronic image stabilization is a technical of making the video sequencesstable, it combines computer technology, electronics technology, digital signalprocessing technology, etc. Electronic image stabilization has many advantages, suchas easy operation, small volume, low power consumption, good performance inreal-time, stability, high precision aspect. Therefore, electronic image stabilizationobtained people’s favor in recent years. In electronic image stabilization technology,the most important task is to find the motion vector between the current frame imagewith jitter and the reference frame image, and using the motion vector to makemotion compensation on the shaking current frame image, established a stable frame.Nowadays, the feature method is the most common method, extract feature pointsbetween the shaking current frame image and the reference frame image, matchingthe feature points, to compute the motion vector of the current shaking frame image.In the feature point matching algorithm, the binary feature descriptor has lowercomputational complexity, and the small amount of calculation, and needs smallstorage. Matching using the Hamming distance between binary feature vectors, andHamming distance calculation using xor and counting operation, the execution on themodern CPU can be very quickly, and requires less storage. BRIEF is a kind of rapidand high robust binary descriptor.First, although BRIEF operator shows better performance in more scenarios, butit hasn’t rotation invariance and scale invariance, therefore, in this paper, the BRIEF operator added rotation invariance and scale invariance, and do not reduce otherperformance.Secondly, the matching method of the binary feature descriptor is usuallycalculated by the Hamming distance. Iterate through all the feature points, select theminimum Hamming distance of feature points as the matching point. This paperproposes an improved matching criterion, choose the value of the minimumHamming distance, then select the second minimum Hamming distance, and computethe ratio of the nearest neighbor Hamming distance and the second nearest neighborHamming distance. When the ratio is below a certain threshold, the selected featurepoints is the matching points.Thirdly, in view of interference of the motion of foreground objects in the videoto the global motion of the whole image, the paper proposed a method of dividing theimage into blocks to remove the interference of the motion of foreground objects tothe global motion of the whole image, make motion estimation of global motionvector more accurate.Finally, this paper introduces the evaluation method of electronic imagestabilization, and use the proposed algorithm to do experiments with multiple jittervideos, contrast the videos both before and after the process, and taking a few framesin video to compare and analyze them.