Extraction device and extraction method of area of interest

Abstract

The invention provides an extraction device and an extraction method of an area of interest. The area of interest is extracted from an image at a high precision, and the interest degree of the area of interest is calculated. The extraction device of the area of interest comprises an extraction component, a retrieval component and an interest degree determining component, wherein the extraction component extracts one or a plurality of partial areas from an input image; the retrieval component retrieves an image similar to the partial area from an image database which stores a plurality of images in regard to the respective partial area extracted by the extraction component; and the interest degree determining component determines the interest degree of the respective partial area on the basis of the retrieval result of the retrieval component.

Description

technical field [0001] The present invention relates to techniques for extracting regions of interest from images. Background technique [0002] Conventionally, there are various prior art techniques for detecting (extracting) a region of interest (an image region expected to be noticed by a person or an image region to be noticed) in an image. In addition, the region of interest detection is also called salient region detection (Saliency Detection), objectness detection (Objectness Detection), foreground detection (Foreground Detection), attention detection (Attention Detection), and the like. These prior arts are broadly classified into learning-based algorithms and model-based algorithms. [0003] In the learning-based algorithm, a pattern of a region to be detected is learned based on a plurality of image data of a learning object, and a region of interest is detected based on the learning result. For example, in Patent Document 1, it is described that the types of fea...

AI Technical Summary

Problems solved by technology

[0007] In the learning-based algorithm, although there is no need to build a brain response model, there is a disadvantage that the detection result depends on the learning data and cannot detect objects that are not similar to the learning data

On the other hand, in the model-based algorithm, the region of interest can be detected without prior knowledge, but it is difficult to construct a model, and the detection accuracy of the region of interest is insufficient.

Therefore, neither method can extract the region of interest with high precision without limiting the detection object

[0008] In addition, even if it is any algorithm based on learning and model, when multiple regions are detected from an image, it is impossible to judge which region is more important and which people care more about.

Images