Hierarchical Motion Vector Estimation With Decision Criterion On Buffer Memory Status

dbstract- This paper presents a new approach of the motion vector estimation, which saves the total number of data on motion vectors in coded images and produce good quality of reconstructed images. The estimation scheme is based on the hierarchical structure to accelerate the processing and adopts a decision criterion to control the amount of incoming data into transmission buffer memory. The decision factors are the prediction error of motion vector on low resolution images and the status (fullness) of transmission buffer memory. To demonstrate the efficience of the proposed process, the comparison of signal to noise ratio(PSNR) with and without using the proposed decision criterion is observed. SUMMARY Due to the limited transmission channel capacity, the motion vector estimation plays an important role to compress the transmitted sequential image data. The quality of reconstructed image is, therefore, strictly related to the motion vectors and its accuracy. However, besides of the accuracy of motion vector, the status of the transmission buffer memory is another important factor for the image quality. The amount of data representing image is directly controlled by buffer status. In case of that it is unnecessary to assign more bits to the motion vectors, it is encourage to assign the saved bits to image information instead. Therefore, the main focus of this paper is reducing the total amount of motion vector data to help assign more bits to image data considering the accuracy of the motion vector and buffer status. To reduce the number of motion vector data, a two-level resolution structure is proposed. The lower resolution image is generated from the original image by reducing it into half resolution in horizontal and vertical directions, therefore it is one quarter size of original image. Gaussian pyramid coding technique 111 is used to reduce the resolution because of its fast performance of removing the spatial redundancies of the images. Motion vectors on low resolution images, with the same size of block and search area in higher resolutions, are estimated. From this coarse motion vector estimation, total number of motion vectors are reduced by factor of four and actual estimated area expands four times larger than one in higher resolution. In other words, one motion vector in lower resolution representsfour motion vectors in higher resolution and the scale of motion vectors detected in low resolution represents twice larger scale in higher resolution. For the accuracy of motion vector, the Block Matching Algorithm (BMA) is used. In BMA, the motion vector is estimated by minimizing the prediction error. The Prediction error of vector (cw,B) is defined as follows ;