摘要:
Each of element processors arranged in correspondence to pixels of a template block and a search window block respectively includes an A register and a B register provided in parallel with each other for storing search window block pixel data respectively, and a T register for storing template block pixel data. Motion vector evaluation value calculation is performed through a first one of the A and B registers and the pixel data stored in the T register, while operated data is transferred to the second one of the A and B registers from the first one of the A and B registers in parallel with the calculation operation, for storing head search window block pixel data of a next search window. A motion vector is detected at a high speed in excellent coding efficiency.
摘要:
A processor array includes element processors which are arranged in a matrix in correspondence to respective pixels of a template block, which is a current picture image pixel block. Each element processor stores pixel data of a search window block, which is a corresponding reference picture image pixel block, and obtains an evaluation function value component with respect to the template block pixel data. A summing part sorts the evaluation function components received from the respective element processors of the processor array in accordance with a plurality of predictive modes and sums up the components for the respective sorts, for forming evaluation function values for the respective predictive modes. A comparison part compares the evaluation function values received from the summing part for each predictive mode, to decide a displacement vector providing the best similarity as a motion vector for each predictive mode. It is possible to simultaneously detect motion vectors according to a plurality of predictive modes. It is possible to detect motion vectors employed for moving image predictive compensation in accordance with a plurality of predictive modes at a high speed with a small hardware volume.
摘要:
An evaluation value operation part computes evaluation values of a template block and a search window block in accordance with respective ones of a plurality of predictive modes in parallel with each other, and a candidate vector determination part decides candidate vectors indicating optimum vectors in accordance with the computed evaluation values and on the basis of priority levels from a priority generation part. In accordance with these candidate vectors, an optimum vector decision part decides the optimum vectors for the respective predictive modes. Thus provided is an image coding system which can reduce the amount of codes of motion vectors with excellent picture quality.
摘要:
Disclosed is an image coding method and apparatus for preventing a coding quantity of coding data from being increased when a redundancy between image planes is low. A selector outputs either an optimum motion vector output from a motion vector detecting device or a motion vector output from a motion vector storing section to a real time image coding device for performing coding on the basis of an evaluation value output from the motion vector detecting device. If it is decided, according to the evaluation value, that a redundancy between a reference image plane and a coding object image plane is low, the motion vector is selected so that the coding quantity of the coding data can be prevented from being increased.
摘要:
A plurality of motion detecting units of which at least a template block size, i.e., the number of estimation pixels, or a search area size is different among the motion detecting units are used adaptively according to the characteristic of, or the prediction coding type of, a target picture. This enables efficient motion vector detection to be accomplished without increasing the amount of hardware or power consumption.
摘要:
Screen data consists of two sets of field data. Each set of field data is divided into a plurality of data blocks which has four rows of pixel data corresponding to four rows of pixels vertically arranged. Every data block corresponding to one set of field data is stored in the first bank (bank0) of a frame buffer memory while that corresponding to the other set of field data is stored in the second bank (bank1). One row address is assigned to each data block. Bank1 is precharged while bank0 is in a write operation and vice versa in order to carry out the precharging operation and the write operation concurrently, so that the pixel data can be transferred at a high data transfer rate and each of two sets of field data can be transferred independently.
摘要:
Instructions of a program are stored at addresses sequentially designated in accordance with an M series pseudo-random number sequence in an instruction memory in the order of program addresses. A pseudo-random number program counter has a feedback shift register for generating the same M series pseudo-random number sequence and applies an address of an instruction to be read from the instruction memory to the instruction memory based on a generated pseudo-random number, and a jump address and a select signal from an instruction decoder. As a result, instructions are read from the instruction memory and executed in the order of program addresses. The feedback shift register can be implemented as a small-scale circuit and operable at high speed.
摘要:
Element processors (PE00 to PE33) included in a processor array (7) store pixel values of a search window, shifting them forward. Further, only hatched element processors (PE00, PE02, PE11, PE13, PE20, PE22, PE31, PE33) store pixel values of a template block, and compare them with the pixel values in the search to evaluate a similarity of pixel values. In other words, the pixel values of the template block are skipped and the pixel values which are left after skipping are compared. Therefore, it is possible to cut a hardware volume.
摘要:
Introduced in an image data encoding device is a logic-memory combined chip in which a memory device and a signal processing device are combined. A logic part and a memory part can be connected to each other with a wide bus, allowing to improve processing capability of data transfer. However, the memory part of the logic-memory combined chip has a small capacity, so that an attempt to increase the capacity will lead to upsizing of the chip, resulting in an increase in costs. Therefore, in a processing that data transfer results in bottlenecks (i.e., motion search), image data is transferred to/from an internal memory capable of transferring data at high speed. In a processing that high-speed data transfer is not required, image data is transferred to/from an external memory.
摘要:
The control circuit controls output of template block data held in the input section such that a plurality of operation units within the operation section are provided with data of unadjacent template blocks that are different from each other. The operation units within the operation section detect motion vectors according to the template block data provided thereto. Thus, motion vector detection in the template blocks except for those in a region on a display screen requiring no motion vector detection is distributed in the plurality of operation units. Accordingly, the motion vector search can be performed in a wider range in a vertical or horizontal direction than in the conventional case within the same operation time.