Image Processing Device and Semiconductor Device

Abstract

In an image processing device, a motion image decoding processing unit extracts a feature amount of a target image to be decoded from an input stream, and changes a read size of a cache fill from an external memory to a cache memory, based on the feature amount. The feature amount represents an intra macro block ratio in, for example, one picture (frames or fields), or a motion vector variation. When the intra macro block ratio is high, the read size of the cache fill is decreased.

1 - 20 . (canceled) 
     
     
         21 . An image processing method, comprising:
 (a) inputting a stream including motion images;   (b) extracting a feature amount of a decoding target image from the stream;   (c) specifying a read size from an external memory to a cache memory based on the feature amount;   (d) determining occurrence of a cache hit or a cache error by comparing a request address to the external memory and tag data;   when the cache hit has occurred, reading data corresponding to the request address, and supplying the data to a decoding processor; and   when the cache error has occurred, reading data regarding the read size specified in step (c) from the external memory, and writing the data to the cache memory.   
     
     
         22 . The image processing method of  claim 21 ,
 wherein an intra macro block ratio in entire macro blocks of one picture in the target image is extracted as the feature amount, and   wherein the read size is decreased as the intra macro block ratio is high, and the read size is increased as the intra macro block ratio is low.   
     
     
         23 . The image processing method of  claim 21 ,
 wherein a motion vector variation of one picture in the target image is extracted, as the feature amount, and   wherein the read size is decreased as the motion vector variation is large, and the read size is increased as the motion vector variation is small.   
     
     
         24 . The image processing method of  claim 23 , including:
 dividing a reference image corresponding to the target image into a plurality of areas based on a direction and a distance centrally about a target macro block to be decoded, and   measuring a frequency distribution of motion vectors specifying the areas,   wherein the feature amount is extracted based on the frequency distribution.   
     
     
         25 . The image processing method of  claim 21 ,
 wherein the stream is input to a motion image decoding processing unit including a variable length encoding processing unit and an image signal processing unit performing a decoding process for a decoded result of the variable length encoding processing unit,   wherein the variable length encoding processing unit and the image signal processing unit operate with a pipeline in unit of one picture,   wherein the feature amount is extracted from the decoded result of the variable length encoding processing unit, and   wherein the request address corresponds to a request from the image signal processing unit.   
     
     
         26 . The image processing method of  claim 21 ,
 wherein encoding information corresponding to the stream is input to an encoding information processing unit, and   wherein the feature amount is extracted from the encoding information supplied through the encoding information processing unit.   
     
     
         27 . The image processing method of  claim 26 ,
 wherein the stream is generated by performing an encoding process for a motion image, and encoding information in the encoding process is supplied to the encoding information processing unit.   
     
     
         28 . The image processing method of  claim 25 ,
 wherein an intra macro block ratio in entire macro blocks of one picture in the target image is extracted as the feature amount, and   wherein the read size is decreased as the intra macro block ratio is high, and the read size is increased as the intra macro block ratio is low.   
     
     
         29 . The image processing method of  claim 25 ,
 wherein a motion vector variation of one picture in the target image is extracted as the feature amount, and   wherein the read size is decreased as the motion vector variation is large, and the read size is increased as the motion vector variation is small.