Image Retrieval Algorithm Based Image Intelligent Design

Adobe Photoshop software with its powerful image editing capabilities is widelyused in wedding photography design, advertising design and many other imageprocessing applications. The PSD is the specialized image format of Photoshop, and itcan save the Photoshop layers, channels and many other image information. But usingPhotoshop for image editing and designing costs a lot of time and manpower. With theexplosive growth of image data, CBIR (Content-based Image Retrieval) has become aresearch hotspot in the field of image processing. And it has important significanceand practical value in the area of computer vision, image processing, databases,artificial intelligence and many other fields. In this paper, PSD file format and imageretrieval algorithm have been researched. In the meantime，this paper combinescolor-based image retrieval algorithms, and it creatively constructs an intelligentimage design system which uses image retrieval algorithm as its core. This intelligentimage design system not only can achieve a variety of image effects which areconsistent with Photoshop, but also can reduce the manual intervention significantly.This paper is as follows:1. This paper systematically analyzes the structure and role of the PSD files,including PSD file header, color mode data, image resource data, layer mask data andimage synthesis data.2. The information extraction principles and methods of PSD files are describedin detail. According to the different characteristics between JSON and XML, JSON ischosen as the project’s basic data description format. This paper also proposes amulti-layer image synthesis method. The overall architecture of the software isintroduced by the actual cases. Finally, this paper describes the contents of each partof the software, and it shows the basic interface software and the using results.3. The intelligent design part of the project is discussed in the last section of thispaper. HSVG one-dimensional non-uniform quantization scheme which is based onHSV color space is proposed. Combining membership colors statistical methods andusing the quantized color histogram as the basic characteristics of the image, it solvesthe color quantization problem caused by the fixed threshold. Finally, hue enhancedhistogram intersection method and the improved quadratic distance algorithm arebeing proposed. The capabilities of these algorithms are compared in three aspects including the objective performance, subjective effects and the calculation speed.