To provide an image data processing unit that can correct a quantization error of an image data component after inverse conversion into prescribed frequency conversion without the need for troublesome processing.
A range over pixel check means 41 seeks a pixel having a component in excess of an upper limit of a range and a component correction means 42 decreases the component value of the pixel so as to be kept within the range. The decreased value can be compensated by deciding a pixel in advance and increasing the same component of the pixel through the use of a stochastic magnification correlation of a product between each element of a matrix used for the inverse conversion and each quantization error caused in a frequency conversion coefficient multiplied with each element by means of a component value distribution means 43.
Next Patent: IMAGE CODER AND IMAGE DECODER