Abstract: A statistical value calculation part specifies macroblocks positioned around an object macroblock and calculates a minimum average value of activities of the macroblocks. When images of the macroblocks are flat and the minimum average value is smaller than an activity of the object macroblock, the minimum average value is set as an adjustment value. A correction factor determination part determines a correction factor on the basis of the adjustment value and a factor determination table. By multiplying a reference quantization step value by the correction factor, a quantization step value of the object macroblock is determined. Since the quantization step value reflects a distribution of the activities of the macroblocks, it is possible to suppress a local change of the quantization step value.

Abstract: It is an object to provide a method of calculating a coding cost by which the magnitude relation of the amounts of generated codes can be estimated with high accuracy. A cost calculation part generates a differential block between a coding object block and a prediction block. Hadamard Transform is performed on the differential block to generate a frequency component block. A conversion factor matrix is generated with the information of a quantization matrix reflected thereon. A coding cost is calculated by multiplying components in the frequency component block individually by components in the conversion factor matrix and adding up the multiplied components. A mode selection part selects an optimum predictive coding method on the basis of the coding cost.