Abstarct: Unprofessionally captured images usually suffer from unwanted vibrations and vibrations of the camera sensor, and usually due to the shaky motion of an unstable handheld camera or a camera mounted on the car, causes noise in the recorded video, which makes the viewer feel annoyed. Today, different methods are used to stabilize video images, and increasing the accuracy of stabilizing video images and also reducing the model error is one of the most important challenges in this area, so in this study, a combined method is used to stabilize video frxames. In this model, first the image content recognition process is performed using Grow Cut method and Zernike Algorithm and then a block matching method is used to stabilize the image. In fact, the image stabilization process in the proposed method consists of two parts: image content detection and image stabilization. In the first part, the image content recognition process is performed. In fact, the image content recognition process consists of two parts, transferring the frxame to Grow Cut and extracting the feature using the Zernike algorithm. In the second part: the image stabilization process is performed. In this part, a block matching method is used to stabilize the image. In order to evaluate the proposed model, the parameters of variance, mean square displacement and Global Motion Vector(GMV) parameters are used and values of GMV those used to calculate other parameters normalized between 0-1. The results of this study show that the value of variance of the proposed model in the X axis is equal to 0.0516 and the amount of variance of the model in the Y axis is equal to 0.0455, also the amount of average image movement in the X and Y axes baxsed on GMV figures in the optimal value The GMV on the X-axis in the proposed model is 0.4509 and for Y-axis amont of 0.4161 reported. This research faced with some limits like lack of computational parameters and most of the comparation between the results should apply by human judgements from watching the final video. To overcome this problem, we used the mean square displacement measure. It represents the difference of displacement of input video with, proposed method and basic method [80]. And the results of that shows amount of 0.0708 for X-axis and 0.0867 for Y-axis between input and baxse method, and results equal to 0.3281 for X-axis and 0.133 for Y-axis between input and proposed method. As long as this parameter measures the displacement between input video that has most vibration and displacements, the results of the best output should be more that the other. So it shows that proposed method did much better in order of mean square displacement measure. Beside that in this reseach there was no dataset or much used input videos, and the input video created for that by merging two input videos used by other works in this research area.