[[["容易理解","easyToUnderstand","thumb-up"],["確實解決了我的問題","solvedMyProblem","thumb-up"],["其他","otherUp","thumb-up"]],[["缺少我需要的資訊","missingTheInformationINeed","thumb-down"],["過於複雜/步驟過多","tooComplicatedTooManySteps","thumb-down"],["過時","outOfDate","thumb-down"],["翻譯問題","translationIssue","thumb-down"],["示例/程式碼問題","samplesCodeIssue","thumb-down"],["其他","otherDown","thumb-down"]],["上次更新時間:2025-08-08 (世界標準時間)。"],[[["\u003cp\u003eThis superseded study compared WebP, JPEG, and JPEG 2000 compression using a dataset of 1 million images, primarily JPEGs.\u003c/p\u003e\n"],["\u003cp\u003eWebP achieved an average compression percentage of 41.30% with non-negative compression gain, outperforming JPEG and JPEG 2000.\u003c/p\u003e\n"],["\u003cp\u003eJPEG 2000 had lower compression percentages compared to JPEG in some cases due to negative compression for many images.\u003c/p\u003e\n"],["\u003cp\u003eWebP provided particularly significant file size reduction for smaller images, which are most common on the web.\u003c/p\u003e\n"],["\u003cp\u003eThe study used an earlier version of WebP and a new study with lossless images as source supersedes this one.\u003c/p\u003e\n"]]],["The WebP team compared WebP, JPEG 2000, and Re-JPEG compression methods on a dataset of 900,000 JPEG images. Each method targeted a specific peak signal-to-noise ratio (PSNR), with WebP achieving an average PSNR of 39.38. Results indicated WebP provided higher overall compression, particularly for smaller images. WebP achieved an average compression of 41.30% with no negative compression. JPEG 2000 averaged 27.67%, and Re-JPEG 22.37%, showing WebP as the superior method.\n"],null,["# Comparative Study\n\n*Comparative study of WebP, JPEG and JPEG 2000, September 2010*\n\n**Note:** We had conducted this initial study with the first version of WebP\nthat we released, which used the `libvpx` library. In addition, the source\nimages were JPEG images that we crawled from the Web. Based on feedback from\nthe community, we have now undertaken a\n[new study](/speed/webp/docs/webp_study) with version 0.1.2 of `libwebp`\nstarting with lossless images (PNG) as source. The new study supersedes this\nstudy.\n\nThe WebP team performed a comparative study of three image compression methods\nover a random set of 1 million images. The goals of the study were to measure\nthe compression achieved by those methods and to analyze the trade-offs\nbetween image sizes and compression.\n\nThree compression methods, WebP, JPEG 2000 and Re-JPEG, were applied to the\n900,000 JPEG images contained in the data set. JPEG images were re-compressed\nwith Re-JPEG so that each was as close as possible to a target peak\nsignal-to-noise ratio (PSNR) value. The team's analysis was done on the images\nafter they had been compressed.\n\nIn addition to the large scale study, the team manually inspected 100 random\nimages and found that, in most cases, WebP provides acceptable visual quality.\n\nConclusions\n-----------\n\nWebP achieves overall higher compression than either JPEG or JPEG 2000. Gains\nin file size minimization are particularly high for smaller images which are\nthe most common ones found on the web.\n\nData Set\n--------\n\nThe data set used in the study was a collection of 1 million images randomly\nsampled from a repository of images crawled from the web. The following table\nshows the distribution of different types of images in the data set.\n\n| Type | Count |\n|-------|--------|\n| JPEG | \\~900K |\n| PNG | \\~47K |\n| GIF | \\~47K |\n| Other | \\~6K |\n\nExperiment Design\n-----------------\n\nThe experiments were run in two stages:\n\n1. The team ran WebP on the images with a target PSNR of 42. They tuned the\n quality parameter (QP) for the WebP encoder until they achieved a result\n that was as close to 42 as they could get for a given image. The actual\n PSNR achieved for each image, therefore, was not necessarily exactly 42.\n\n2. The team used the PSNR value for the WebP-compressed images as the target\n PSNR for the JPEG 2000 and Re-JPEG images. They also performed a line\n search over the quality parameters of the JPEG 2000 and JPEG images in\n order to come as close as possible to the target PSNR.\n\nThe actual compression result was measured as a percentage using the following\nformula: \n\n compression_percentage = 100 * (original_image.length - compressed_image.length) / original_image.length\n\nResults\n-------\n\nThe tests yielded the following results for negative and non-negative\ncompression gain:\n\n### Non-negative Compression Gain\n\nThe team did not penalize a method if that method increased the image size\nafter compression. That is, the compression_percentage was considered to be 0\nwhen compressed_image.length was greater than the original_image.length.\n\n### Negative Compression Gain\n\nThe team allowed negative compression percentage results to get a better\nunderstanding of the various compressors' raw performance.\n\nFollowing are the average compression percentages that were achieved by each\nof the three methods.\n\n| Type | Avg PSNR Obtained | Avg Compression % (non-negative compression gain) | Avg Compression % (negative compression gain) |\n|-----------|-------------------|---------------------------------------------------|-----------------------------------------------|\n| WebP | 39.38 | 41.30 | 39.80 |\n| JPEG 2000 | 39.49 | 27.67 | 9.71 |\n| Re-JPEG | 39.36 | 22.37 | 14.62 |\n\nWhen negative compression gain was allowed, the average performance for JPEG\n2000 was lower than that of JPEG. This may be due to high-frequency artifacts\npresent in the JPEG images; these artifacts can contribute to a\nless-than-ideal compression situation.\n\nThe Kakadu implementation used in the experiment to convert to JPEG 2000\nfailed to convert approximately 240,000 images.\n\nJPEG 2000 had lower compression percentages than Re-JPEG, primarily because\nthe JPEG 2000 result was negative compression for many images. For that\nreason, JPEG images achieved a lower overall compression score.\n\nThe following figures provide distribution graphs for the three methods:\n\n**Figure 1: Image Size Population Distribution**\n\nThe graph shows that most of the images were small. The frequency of the\nimages with a size more than 500K is generally less than 100.\n\n**Figure 2: Compression Percentages for WebP, JPEG, and JPEG 2000 Images**\n\nThis graph demonstrates that the body of pictures compressed with WebP got\nbetter compression than re-jpeg and jpeg 2000. Also, the pictures compressed\nwith WebP got more positive compression than the others.\n\n**Figure 3: Comparison of Image Sizes and Compression Percentage**\n\nThis graph shows that WebP achieves better compression than the other formats,\nparticularly for smaller images."]]