Image interestingness dichotomy prediction method combining discriminant analysis and multi-kernel learning

A technology of multi-core learning and discriminant analysis, applied in the field of image analysis, which can solve problems such as redundancy of interesting features

CN110569860AActive Publication Date: 2019-12-13南京鹰视星大数据科技有限公司
3 Cites 5 Cited by

Patent Information

Authority / Receiving Office
CN · China
Current Assignee / Owner
Publication Date
2019-12-13

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses an image interestingness dichotomy prediction method combining discriminant analysis and multi-kernel learning, and the method comprises the steps: inputting image data, and forming a data set; inputting the data set in the step 1, and determining three clues including an unusual clue, an aesthetic clue and a general preference clue in the data set; carrying out any featurefusion by adopting discrimination correlation analysis or multiple discrimination correlation analysis; and a simple multi-kernel learning algorithm is adopted for classification. According to the image interestingness dichotomy prediction method, compact expression of different interestingness characteristics in each clue and interestingness multi-source heterogeneous characteristics of expression between the clues are considered, a compact and discriminative interestingness characteristic set is formed, and simultaneous characterization and modeling of multi-source interestingness information are realized.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention belongs to the technical field of image analysis, and in particular relates to a binary classification prediction method for image interestingness combined with discriminant analysis and multi-kernel learning. Background technique

[0002] In recent years, with the increasing number of users of various portal websites and social platforms, large-scale and massive image data continue to emerge, which poses increasing challenges to image retrieval systems that can meet user preferences. Among them, the interestingness of images is a major type of user preference, which requires the image push platform to provide image data that meets the characteristics of interestingness according to the user's query goals, and at the same time meet the semantic and emotional expectations of users. At present, the existing image interestingness binary classification methods mainly focus on the exploration and simple and direct utilization of image interest...

Examples

Embodiment Construction

[0142] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0143] The present invention combines discriminant analysis and multi-kernel learning image interestingness binary classification prediction method, such as figure 1 shown, including the following steps:

[0144] Step 1: Input image data to form a data set;

[0145] The data set used in the present invention is the data set provided in the Predictive Multimedia Interesting Task Competition released in 2016, which consists of Hollywood movie trailers licensed by Creative Commons. The entire data set includes 78 trailers in total. The corresponding trailers are divided into video shots, and the middle frame of each shot is taken as the image data. The entire data set has 7396 images in total, and the present invention divides the entire data set into a training set and a test set according to a ratio of 7:3.

[0146] For the annotation of im...