Abstract: An image processing method whereby data pertaining to an estimated captured image obtained from reference data of a sample is acquired using an input acceptance unit, an estimation unit, and an output unit. The data is used when comparing the estimated image and an actual image of the sample, wherein the method includes: an input acceptance unit accepting input of the reference data, process information pertaining to the sample, and trained model data; the estimation unit using the reference data, the process information, and the model data to calculate captured image statistics representing a probabilistic distribution of values attained by the data of the captured image; and the output unit outputting the captured image statistics, and generating the estimated captured image from the captured image statistics. This permits reducing the time required for estimation and to perform comparison in real time.

Abstract: A learning processing device and method achieves learning of a lightweight model that is completed in a short amount of time. The learning processing device obtains a new, second learning model from an existing first learning model. An input unit acquires a first learning model generated in advance by learning a first learning data set, and an unpruned neural network (hereinafter, NN). An important parameter identification unit uses the first learning model and the NN to initialize a NN to be learned, and uses a second learning data set and the initialized NN to identify a degree of importance of parameters in a recognition process of the initialized NN. A new model generation unit carries out a pruning process for deleting parameters which are not important from the initialized NN, thereby generating a second NN; and a learning unit uses the second learning data set to learn the second NN.

Abstract: This disclosure relates to a system for performing efficient learning of a specific portion. To achieve this purpose, there is proposed a system configured to generate a converted image on the basis of input of an input image, the system comprising a learning model in which parameters are adjusted so as to suppress an error between the input image and a second image converted upon input of the input image, the learning model being subjected to different learning at least between a first area in the image and a second area different from the first area.

Abstract: The purpose of the present invention is to provide a computer program for achieving die-to-database inspection at high speed and with few false reports, and a semiconductor inspection device using the same. To achieve this purpose, the present invention proposes: a computer program comprising an encoder layer that is configured to determine the features of a design data image, and a decoder layer that is configured to generate, on the basis of a variation in an image (inspection target image) obtained by photographing an inspection target pattern, a statistic pertaining to the brightness values of pixels from feature values output by the encoder layer, wherein die-to-database inspection with few false reports can be achieved by comparing the inspection target image and the statistic obtained from the decoder layer and pertaining to the brightness values, and thereby detecting a defect region in the image; and a semiconductor inspection device using the same.