摘要 |
A digitally inputted image signal (SD signal) is converted into a high resolution digital image signal (HD signal) by using estimated values. The specific pixels of an object to be estimated are classified according to the one-, two-, or three-dimension level distribution of plural reference pixels of SD signal located near the specific pixels. The estimated values of the specific pixels are generated by a linear combination of the values of the reference values and the estimation coefficients determined in advance by learning. During the learning, a known HD signal and the SD signal formed from it are used to determine the estimation coefficients so as to minimize the sum of squares of the errors between the true values and the values estimated by the linear combination of the values of the surrounding SD signal reference pixels and the estimated coefficients. Not necessarily limited to the estimated coefficients, it may be possible to use, as the estimated values, representatives determined for every class in correspondence with the class of the inputted SD signal. The reference value of a block and the value normalized by the dynamic range DR are an example as the representatives. <IMAGE> |