Electronic Image Stabilization Algorithm Based On The DM642

Electronic image stabilization (EIS) technology‘s core processing unit is DSP chip, inwhich digital image processing algorithm was used to remove irregular image shake andobtain stable image. The motion estimation of the current frame is the first step, in which thedisplacement value of the current frame is calculated, and the current frame’motioncompensation is the next step. Compared with mechanical image stabilization, optical imagestabilization and other image stabilization techniques required complex hardware supporting,the algorithm only needs to changing in electronic image stabilization. Image stabilization’saccuracy and speed depend on algorithm itself, electronic image stabilization is a hot area ofimage stabilization.In the paper, several basic electronic image stabilization algorithms and the step ofrealizing electronic image stabilization on the DM642 platform were described in detail. Tothe small translation image shake, the structure of capturing two images was constructed inthe non BIOS project, and the SAD norm was corrected properly in the odd even field, thenthe 30×30 small blocks were selected in the image, and two images registration was realizedin the 30 pixels range. In the BIOS project, the 60×60 small blocks were selected in theimage, and two images registration was realized in the 30 pixels range. To the largetranslation image shake, the structure of capturing twenty four images was constructed andgray scale projection algorithm was applied in the BIOS project, and the match of twenty fourimages was realized in the range of 60 pixels. The frist adjacent frame was selected as thereference frame and the second to twenty four adjacent frames as the current frames, and thedisplacement values of current frames were got in the method of calculating the row columncross correlation operation value, at last the current frames were compensated in the way offixed frame compensation. To the rotation image shake, harris feature point detection algorithm was applied in the BIOS project, and two images match was realized. In thealgorithm, 320 feature points were selected in the reference frame, and the correspondingfeature points were searched in the current frame. The current frame’s motion parameter wascomputed in the method of exploiting the position relationship of two group feature points, atlast the current frame was compensated.In experiment, the shaking image sequences were obtained by shaking camera, and thenthe algorithm was compiled in the matlab and planted on the DM642, so that the shake imagewas compensated. The PSNR value was improved to certain extent after the shaking imagesequences was compensated with three different algorithms. The local image information isexploited in block match algorithm, but the whole image information is exploited in grayscale projection algorithm. To the translational shake, gray scale projection algorithm is betterthan block match algorithm in stabilization range, precision and speed in theory, and theexperiment results coincided with theory. To the rotation shake, the shake of image sequenceswas also removed to certain extent after harris feature point algorithm was utilized.