Activities are normalized corresponding to a characteristic of a pattern of a frame. The picture quality is optimized by adaptive quantization. A normalized activity norm_act [m] of each macro block is obtained with an average activity avg_act of the frame and the activity act [m] of each macro block according to the following formulas (1) and (2). In the following formulas, att is a parameter, for example att=0.125. norm_gain=att.times.avg_act+1 (1) norm_act[m]={(norm_gain.times.act[m].times.act[m])+(avg_act.times.avg_act- )}/{(act[m].times.act[m])+(norm_gain.times.avg_act.times.avg_act)} (2)Thus, the normalized activities norm_act [m] are normalized in the range from 1/norm_gain to norm_gain. Since norm_gain is proportional to the average activity avg_act, when a picture has a flat pattern whose average activity avg_act is small, the normalizing range becomes small. Thus, the quantized values of macro blocks are not largely different. As a result, the picture is equally quantized. In contrast, when a picture has a complex pattern whose average activity avg_act is large, the normalizing range becomes large. Thus, the quantized values of macro block are largely different. As a result, a macro block having a flat pattern is finely quantized, whereas a macro block having a complex pattern is coarsely quantized.